AI | Freedom.Tech

Ollama 0.33.1

Ollama 0.33.1 adds Qwen3.8 Flash support, structured output for MLX models, and fixes Metal GPU timeouts during model loading

Anthropic Sdk Ruby 1.66.0

Adds thinking display mode beta, organization usage support, discriminated union module enums

Mistral Vibe 2.24.4

Mistral Vibe adds LLM-generated session titles, queued message editing, reused model connections for speed, and fixes auth routing, rate-limit retries, and connection pooling issues

Hugging Face Transformers 5.16.1

Adds GLM model support

Qwen Code 0.22.2

Qwen Code 0.22.2 adds temporal review lenses, persistent daemon session model selection, MCP 2026 protocol support, and Computer Use SDK v0.20.0 with stable element tokens

Cli 1.1.0

Install elevenlabs-cli 1.1.0 Install prebuilt binaries via shell script curl --proto ' =https ' --tlsv1.2 -LsSf | sh Install prebuilt binaries via powershell script powershell -ExecutionPoli

llama.cpp b10636

Ui: disable the npm UI build (LLAMA BUILD UI=OFF)

Hugging Face Transformers Release 5.16.0

Adds Qwen4-Exp model with GatedResidual, Qwen Sparse Attention, and Per-Layer Embedding components

vLLM 0.28.0

584 commits from 270

Cline SDK 0.0.80

Cline SDK fixes process crashes on huge files, redacts git credentials from prompts, adds seven MCP providers, and updates model defaults across multiple vendors

Cline 4.1.16

Cline 4.1.16 fixes cost estimate display for subscription providers, workspace hook resolution across multiple VS Code windows, credential leakage in model context, and event log disk exhaustion

Qwen Code 0.22.2-preview.1

Preview.1 adds explicit user opt-in for workflows, containerizes code review execution, and improves workflow safety with pre-spend validation

InvokeAI 6.14.0

InvokeAI 6.14.0 adds video generation, supports Flux.2/Krea.2/Ernie/Ideogram models, enables multi-GPU and Intel XPU, and ships FP8 quantization

Claude Code Action 1.0.206

Use paths in delete files prompt example; Allow parentheses in valid branch names; Bound download job log against a stalled log fetch

Claude Agent Sdk Typescript 0.3.246

Claude Agent SDK adds message tracing, cost basis reporting, managed pricing controls, and per-task interrupt handling for multi-agent workflows

Open WebUI 0.11.1

Adds human-in-the-loop tool approval

llama.cpp b10625

Scopes Qwen3-coder chat workarounds

Maple 3.3.8

Agent Mode upgraded to Goose 1.47.0, desktop updates route through protected Maple endpoint, and updater verification hardened

Rslearn 0.1.14

Enhances EarthDaily integration with asset key resolution, per-pixel regression metrics bugfix, logger config saving fix

Packages React 1.14.0

Adds external agent disconnected client event support

Dify 1.17.0

E2B cloud sandbox backend for agent shell and code execution, home snapshots, skill management

llama.cpp 0.3.0

Llama.cpp 0.3.0 adds dots3-note multimodal support, tensor-split for DeepSeek 4, multi-token prediction for GLM-4.5-Air, and WebP image decoding

Claude Agent Sdk Typescript 0.3.245

Claude Agent SDK TypeScript synced to Claude Code v2.1.245, tracking latest model capabilities for agent development

Qwen Code 0.22.0-nightly.20260825.22bb5e8b9f

Nightly build fixes shell session workspace handling, review comment logic, and autofix budgeting, plus adds token accounting and dynamic workflows

qwen-mm-plugins-core 1.0.5

Added NIfTI volume visualization and 4D volume selection support for medical imaging workflows

HelixML 2.12.5

HelixML adds OpenRouter provider support, project-scoped artifact hosting, editable model enforcement, trigger-based workflows, and mobile UI fixes

Cli 1.0.0

Install elevenlabs-cli 1.0.0 Install prebuilt binaries via shell script curl --proto ' =https ' --tlsv1.2 -LsSf | sh Install prebuilt binaries via powershell script powershell -ExecutionPoli

llama.cpp b10610

Metal flash-attention tuning: 53 new f16 instantiations, per-device dispatch tables

Elevenlabs Swift Sdk 3.3.0

Client-error parsing, type-safe tool results, realtime-thread allocation dropped from audio processor

NanoClaw 2.3.0

Breaking change: classic single-bot Slack installs must migrate to per-agent provisioned apps

Cli 1.0.0-alpha.3

Alpha.3 Install prebuilt binaries via shell script curl --proto ' =https ' --tlsv1.2 -LsSf | sh Install prebuilt binaries via powershell script powershel

OpenClaw 2026.8.1-beta.3

GPT-5.6 reasoning support, Puppeteer CDP relay, Gateway lifecycle supervision, and SQLite backup restore now available in OpenClaw and Codex runtime

Qwen Code cua-driver-rs 0.20.0

Prebuilt binaries for Qwen CUA Driver

Qwen Code 0.22.0-nightly.20260824.3a1f86d805

Qwen Code fixes shell session context passing, review lenses, and autofix repair budgets while adding token accounting and dynamic workflow support

llama.cpp b10603

GLM-4.5-Air MTP support added to llama.cpp for improved model inference efficiency

Cline 4.1.14

Cline 4.1.14 updates model catalog with Claude Fable 5, Grok 4.6, DeepSeek V4 variants, and fixes telemetry for interactive session completion

Qwen Code 0.22.0

Web shell now bounds transcript retention and trims oversized replays to prevent out-of-memory crashes

Claude Agent Sdk Typescript 0.3.241

Claude Agent SDK TypeScript now matches Claude Code v2.1.241 capabilities

Qwen Code 0.22.0-nightly.20260823.1007bcacfc

Qwen Code adds test validation for code fixes, temporal incident replay in reviews, and daemon session model persistence with shell and workflow parsing improvements

llama.cpp b10588

Fixes clang LTO build issue

Maple Proxy 0.3.2

Maple Proxy v0.3.2; All binaries include

Cline 4.1.13

Cline 4.1.13 restores tool calling for custom OpenAI-compatible models, fixes Hub session persistence across restarts, and adds session identity to Langfuse traces

llm CLI 0.33

Migrates to OpenAI Python library 3.x, switches HTTP client from httpx to httpx2

Qwen Code 0.21.14-nightly.20260822.7a4566cb3b

Code review loop now explains settlement failures, web shell bounds transcript retention to prevent crashes, and autofix audits approaches instead of stopping on budget breaches

llama.cpp b10568

CI switches to LLVM OpenMP on Windows; Drops non-redist debug build

Ollama 0.33.0

Claude Desktop integration: toggle individual Ollama models on and off from the menu bar, choose models directly within Claude

Maple 3.3.6

Agent Mode now supports image attachments, Goose runtime upgraded to 1.46.0, Linux package updates require confirmation, and RustSec advisory checks enforced daily

Cline 4.1.12

Fixed enterprise MCP marketplace controls and restored tool calling for custom OpenAI-compatible models with empty capability lists

Design.Md 0.2.0

Design.Md now records OlmoEarth subtitle roles and prevents duplicate version tags during releases

Design.Md 0.1.0

Design.md publishes Skylight and EarthRanger design systems with agent skills framework to npm

Claude Code Action 1.0.199

Fixes GraphQL URL handling, MCP server initialization, and tool allowlist validation in GitHub Actions integration

Claude Agent Sdk Typescript 0.3.239

Claude Agent SDK now correctly applies US inference cost multipliers, fixes system prompt boundaries on multi-cloud providers, and improves background task visibility on reconnect

llama.cpp 0.2.0

Llama.cpp 0.2.0 syncs ggml to 0.21.0, adds backend optimizations for CUDA/Metal/SYCL/OpenCL, and improves quantization kernel performance across hardware

llm CLI 0.32.1

Pins OpenAI Python dependency to pre-3.0 to avoid requiring httpx on fresh installs

opencode 1.18.21

Fixes unknown finish reason handling in model responses and improves Vertex AI multi-region routing; Desktop fixes include persistent file search results during loading

langchain-perplexity== 1.4.1

Perplexity integration fixes message type handling, preserves caller parameters, and updates dependencies for compatibility

Deepseek Harness 0.1.1-rc.2

DeepSeek adapter now uses Files API for image uploads with automatic preprocessing, reusing files to reduce API costs

Hermes Agent 0.20.5

Hermes Agent 0.20.5 bundles 323 PRs with group threads, keyless web tier with free provider rotation, CLI improvements, execution guards, and desktop performance optimizations

Cline Desktop adds durable todos and recurring schedules for agents, fixes checkpoint restore and hook handling, consolidates plugins into marketplace, and refreshes model catalog with eight new providers

Cline SDK 0.0.76

Cline SDK adds model-driven image generation, workspace-scoped task scheduling, improved tool activity observability, and fixes hook execution, command routing, and session status tracking

Qwen Code 0.21.15

Web Shell gains file attachment support with faster streaming, Qwen models expose reasoning toggles, review workflows support resumption, and Git credential persistence enables private repository access

Maple 3.3.5

Maple 3.3.5 adds Word document support, expands Agent Mode with richer ACP sessions and reversible per-project trust controls, and blocks known compromised Rust crates

Anthropic Sdk Python 1.0.0

Breaking change: httpx2 upgrade requires dependency refresh

Mistral Vibe 2.24.3

Slash commands now work during agent execution, Python security updates applied, and connector bootstrap errors properly surfaced instead of silently failing

LocalAI 4.9.0

Authentication now deny-by-default, chat adds context compression, MiniMax-H3 video generation, and improved operator visibility across 146 pull requests

whisper.cpp 1.9.3

Whisper.cpp 1.9.3 syncs upstream GGML improvements including CUDA/Metal optimizations, quantization support, and cross-platform backend enhancements for faster inference

opencode 1.18.19

Native OpenAI and Anthropic passthroughs for Cloudflare AI Gateway, improved rate limiting, and fixed provider authentication display and cost calculation bugs

llama.cpp b10505

Adds dedup-cache-models preset option to server

Claude Code Action 1.0.197

Fixed credential handling incompatibility with GitHub Actions checkout v6+ to prevent unintended credential exposure

langchain-openai== 1.6.0

Langchain-openai adds standard model exception types and improves error handling for unexpected OpenAI API response types

langchain-anthropic== 1.6.0

LangChain's Anthropic integration adds standard model exception types and fixes directory search scope issues

Maple 3.3.4

Maple 3.3.4 adds mid-run request queuing with editing, better Agent prompt history navigation, and improved resilience to inference failures

LlamaIndex 0.14.24

Fixes file-handle leaks in HotpotQA evaluator and preserves oversized leaf nodes in CodeSplitter

Cli 1.0.0-alpha.2

Alpha.2 Install prebuilt binaries via shell script curl --proto ' =https ' --tlsv1.2 -LsSf | sh Install prebuilt binaries via powershell script powershel

Packages React 1.13.0

Adds onContextUsage callback for context usage server events

Ollama 0.32.15

Model metadata caching halves time-to-first-token, parser error handling fixed, Qwen system messages normalized

langchain-core== 1.6.0

Resolves tool schema serialization issues, adds standard exception types, and improves Windows compatibility in langchain-core

Deepseek Harness 0.1.0-rc.8

DeepSeek Harness adds multimodal image support, installable subagents, Windows PowerShell sessions, and fixes image payload failures plus OpenAI gateway compatibility

HelixML 2.12.4

HelixML adds DeepSeek code agent runtime, GPU-optional headless tasks, durable agent cancellation, and improved sandbox resilience across API and CLI

Anthropic Sdk Java 2.55.0

Anthropic Java SDK adds files and memory store APIs, fixes message stream accumulation bugs, and upgrades build tooling to Gradle 9 and Kotlin 2.3

Hugging Face Transformers Patch release 5.15.1

Fixes DFlash and MTP candidate generator issues, resolves image processing on accelerators with Lanczos filter

Kimi Code 0.37.2

Kimi Code 0.37.2 expands subagent detail panels and adds optional multi-tab sidebar navigation in Settings Lab

langchain-openai== 1.5.2

Fixes reasoning item boundaries, adds gateway metadata extraction, and extends token counting support to o-series models

Mistral Vibe 2.24.2

Session picker, runtime log control, and worktree management ship alongside faster startup and session resume for improved AI agent workflows

llama.cpp b10486

Fixes LFM2 image tiling threshold; Cross-platform refactor

llama.cpp 0.1.2

Llama.cpp synced GGML 0.20.2, added integer tokenizer scores, improved CUDA performance on DGX, and fixed xcframework builds

langcrew-agentx== 0.1.10

Sets repository.url for npm provenance; Package.json metadata fix

qwen-mm-plugins-edu-agent 1.0.2

Normalizes local file URIs across capabilities; Cross-platform path fix

qwen-mm-plugins-core 1.0.3

Isolates 3D worker from MCP stdio; Windows process-handle and pipe fixes

Hermes Agent 0.20.4

Hermes Agent 0.20.4 stabilizes 74 merged PRs including desktop UI improvements, bot group-chat fixes, NVIDIA security scanning on skill installs, and SessionDB performance hardening

Qwen Code dsw-eas-tb-smoke-20260818-r2

Qwen improved end-to-end credential refresh handling in code tasks, verified against SWE-bench and Terminal-Bench benchmarks

langchain-openai== 1.5

Adds gateway metadata extraction from response headers. Fixes o-series token counting

Cline Desktop 0.0.14-beta.1

Beta release adds cloud sessions for syncing work across devices, avatar overlay companion, SSH remote execution, and voice input with model-driven image generation

Qwen Code 0.21.11-nightly.20260818.259951c53e

Added session registry with qwen sessions ps, improved web-shell SSE handling and session management, enhanced agent workflow persistence, and added /advisor CLI command for code review feedback

Olmoearth Pretrain Minimal Publish 0.0.7

Removed max_sequence_length constraint to enable flexible sequence handling in pretraining pipeline

Claude Agent Sdk Typescript 0.3.234

Claude Agent SDK TypeScript adds ANTHROPIC_API_KEY source reporting, permission-class message delivery, effort-level tracking, and fixes unused type exports

Hermes Agent 0.20.3

Hermes Agent 0.20.3 bundles MCP 2.x SDK migration, stateless protocol support, Bot Mode plugin, subprocess hardening, and cron scheduler self-healing fixes

llama.cpp b10470

CI; Explicitly pushes release tag before creating GitHub release

Deepseek Harness 0.1.0-rc.7

Deepseek Harness adds durable image attachments in MCP/ACP modes, Job Panel for subagent task management, fixes latency and stack overflows, and introduces configurable reasoning effort levels

Qwen Code 0.21.13

Web shell now accepts text files as named attachments, conversation forking preserves checkpoints, review skill uses platform subcommands, and stopped takeovers get autofix/needs-human labels

InvokeAI 6.14.0

Adds video generation via Wan 2.2; Release candidate 2

Hermes Agent 0.20.2

Hermes Agent v0.20.2 consolidates 397 PRs across desktop, CLI, gateway, and auth with prompt caching, model routing, and platform robustness improvements

koboldcpp 1.119

Video generation and I2V support with Minimax H3

llama.cpp b10452

Removes some ggml concat calls; Model-layer refactor

OpenClaw PR #124528 Gateway profile evidence

CPU profiling data for Gateway PR 124528; Three-node, twelve-turn rig

Qwen Code 0.21.12-preview.5

Upstream release notes did not include a readable summary.

AI release Aug 15, 2026

llama.cpp b10447

Adds Kimi-K3 text-model support with hybrid KDA and MLA attention

AI release Aug 15, 2026

OpenClaw 2026.8.1-beta.2

Binds shared secrets to exact HTTPS destination hosts; Unbound substitution now fails closed before plaintext egress

AI release Aug 15, 2026

Qwen Code 0.21.12

Qwen Code adds drag-and-drop file uploads to Web Shell, implements diff growth limits in autofix reviews, requires executed witnesses for critical findings, and adaptively expands daemon journal capacity to 256 MiB

llama.cpp b10434

Llama.cpp now passes reasoning_effort parameter through chat templates, enabling fine-grained control over model reasoning depth in inference

langchain-openrouter== 0.2.8

LangChain OpenRouter now preserves cost metadata and provider information in usage chunks and response metadata

Elevenlabs Python 2.64.0

Adds extra setup config escape hatch for on-prem deployments

Client Ts typescript 2.6.3

OpenAPI Doc; Speakeasy CLI 1.763.6 (2.884.13); Typescript v2.6.3

Cline SDK 0.0.75

Provider-executed web search, dedicated Cline gateway provider, and Hub daemon stability fixes prevent session crashes during multi-build installations

Qwen Code DSW EAS SWE + Terminal-Bench E2E r2

Qwen3.7-plus model achieves 100% on SWE-bench verified, resolving one software engineering task end-to-end with artifacts and verifier evidence

Ollama 0.32.11

Ollama added DeepSeek Harness and Meta Muse Code agent support, plus web search in OpenAI-compatible Responses API

Qwen Code 0.21.11-nightly.20260814.45c2e73080

Qwen Code adds workspace file uploads, fixes session handling and UI jitter, patched sharp dependency vulnerability, and improves review workflows with witness verification and budget controls

Claude Code Action 1.0.192

Claude Code Action improves GitHub Actions pagination, preserves structured tool output, and detects binary files rather than extension

ComfyUI 0.33.1

Fixes KSamplerAdvanced nested latent handling, adds Minimax Music 3 and Bria image tools, improves VRAM management across platforms

llama.cpp b10419

OpenVINO backend gains Qwen3.5 support, memory optimizations for GPU inference, and fixes for stateful RoPE accuracy and recurrent state handling

AnythingLLM 1.16.0

AnythingLLM adds image generation, mid-session tool toggling, folder drag-drop uploads, and proper generation abortion to stop ghost inference

InvokeAI 6.13.8

Patches two critical security vulnerabilities: arbitrary code execution via crafted URL payloads and unauthorized network connection attempts

NanoClaw 2.2.0

Plugin-stamped agents now update in place with dry-run plans, MCP servers support working directories and HTTP URLs, and templates conform to Agent Plugins 1.0.0 spec for portability

Hermes Agent 0.20.1

Hermes Agent 0.20.1 consolidates 656 merged PRs with stabilization fixes across desktop, gateway, installers, and provider tools

Anthropic Cli 1.23.0

Anthropic CLI adds output_behavior flag to dream creation for choosing between new memory store or in-place input update

Anthropic Sdk Ruby 1.62.0

Output behavior control for memory stores: create new or update in place. Bedrock now uses default-credentials hook

Anthropic Sdk Python 0.122.0

Adds output-behavior control for memory-store creation. Fixes async SigV4 signing in Bedrock clients

HelixML 2.12.3

Fixed MCP proxy resolution, Claude thread recovery, and rate limiter deadlock while adding persistent task terminal, workspace inspection, and inline task editing

Client Python 2.9.3

OpenAPI Doc; Speakeasy CLI 1.763.6 (2.884.13); Python v2.9.3

The builder’s guide to GPT‑5.6

Learn how startups use GPT-5.6 to build faster, more cost-efficient AI agents with smarter model selection and new Responses API capabilities

Qwen Code DSW EAS Harbor full E2E 2026-08-13 r3

Non-production full SWE-bench Verified E2E validation

Cline CLI 3.0.54

Claude Code provider now runs native tools with workspace anchoring and permission rules, tool-call JSON validation rejects malformed payloads instead of silently repairing them, and Hub daemon upgrades directionally without interrupting sessions

Cline SDK 0.0.74

Claude Code provider now declares native tools and respects workspace boundaries and user permissions, tool-call JSON corruption fixed, Hub daemon upgrades directionally, and telemetry now accurately reports token usage without redundant counting

Kimi Code 0.36.0

Experimental fullscreen TUI mode, configurable model pools for subagents, LaTeX rendering, improved interrupt handling, and workspace trust enhancements

Qwen Code 0.21.11

Qwen Code adds Agent Plugins v1 support, multi-agent workflows via /coordinate, and OpenTelemetry observability for session lifecycle tracking

langchain-anthropic== 1.5.6

LangChain Anthropic 1.5.6 fixes tool result block normalization and corrects model profile data for recent Claude versions

Ollama 0.32.10

Ollama 0.32.10 changes default repeat penalty to 1.0 for faster speculative decoding, speeds NVFP4 prefill%, and fixes blob verification on shared digests

Claude Agent Sdk Typescript 0.3.229

SDK now exposes terminal slash commands to clients and properly surfaces request-body-size limits as API errors instead of image errors

opencode 1.18.17

Session compaction now preserves complete recent turns and produces clearer summaries, plus fixes for PDF attachments, MERGE Gateway reasoning, and model routing

Asta Bench Leaderboard Design review attachments

Rendered review evidence hosted outside the source tree and referenced from pull-request metadata

Maple Proxy 0.3.1

Maple Proxy v0.3.1; All binaries include

Kimi Code 0.35.0

Tasks panel now displays live subagent progress, image/video results open in fullscreen with zoom, token counting fixed, and startup freezes from search index loading eliminated

llama.cpp b10369

Build b10375 tightens bare function parsing for Qwen models

Qwen Code 0.21.10-nightly.20260812.a64d1291d2

Qwen Code fixes web-shell session navigation, bounds daemon buffers, improves keyboard interactions, and adds Agent Plugins v1 support

Maple Proxy 0.3.0

Maple Proxy v0.3.0; All binaries include

ComfyUI 0.32.0

ComfyUI 0.32.0 adds LTX 2.5 support, optimizes memory handling for H3 models, and improves nested tensor debugging for complex workflows

langchain== 1.3.15

LangChain 1.3.15 adds state_schema to wrap_tool_call, exposes trace_policy on AgentMiddleware, and fixes middleware control flow, tool call handling, and structured output parsing

Cline CLI 3.0.53

CLI now detects stale daemon upgrades, fixes reasoning model compaction, adds Fable 5 to Vertex catalog, and routes custom model IDs correctly

langchain-core== 1.5.4

Langchain-core 1.5.4 fixes Pydantic 2.14 compatibility, stops StructuredPrompt from mutating caller kwargs, and preserves tool argument schemas

Anthropic Sdk Go 1.63.0

Go SDK adds output_behavior control for dream memory creation, fixes agent toolset path handling and Vertex auth wrapping

Mistral Vibe 2.24.1

Mistral Vibe 2.24.1 adds inline completions, improves agent resume performance, fixes MCP server error handling and voice transcription visibility, and patches security scoping for sensitive file access

Ollama 0.32.9

Ollama adds NVIDIA's Nemotron 3.5 Lightning, a 30B mixture-of-experts model optimized for always-on agent deployment

llama.cpp b10359

Switches ROCm target to 7.14, the first production release using TheRock build system

vLLM 0.27.1

VLLM adds support for quantized DSpark Markov heads, enabling efficient inference with compressed attention mechanisms

Client Ts typescript 2.6.2

OpenAPI Doc; Speakeasy CLI 1.763.6 (2.884.13); Typescript v2.6.2

Ollama 0.32.8

Muse Glimmer model now available across all platforms, optimized for coding agents and personal assistants with MLX performance on Apple Silicon

vLLM 0.27.0

561 commits from 242; Review the changelog before upgrading production inference workloads

llama.cpp b10344

Added MTP (Multi-Token Prediction) support for Nemotron models, enabling faster inference for this architecture

Qwen Code 0.21.9

Qwen Code adds plugin installation from multiple sources, Local Control pairing via QR code, drag-and-drop image support, and improves code review performance for large diffs

Ollama 0.32.7

Muse Glimmer, Meta's 30B multimodal agent model, now runs locally on Ollama with Apple Silicon optimization and image input support

Hugging Face Transformers Release 5.15.0

Adds Meta Muse Glimmer model support for agentic multimodal workflows

opencode 1.18.16

Config parsing now tolerates unknown fields, project registration works correctly from Home, and right-click project menu opens in Desktop mode

Qwen Code 0.21.8

Qwen Code 0.21.8 restores fork PR autofixes, enables compression cache sharing across OpenAI/Gemini/Vertex AI, and improves /review performance to 93.3% cache hit rates

AI release Aug 9, 2026

Maple 3.3.2

Agent Mode gains Goose and RMCP 3 support with improved reliability, sidebar UX, model memory, and Gemma 4 thinking integration

AI release Aug 9, 2026

Cline 4.1.7

Adds task change review via SDK checkpoints, pre-registered OAuth for remote MCP servers, and fixes session durability across aborts, LiteLLM routing, and MCP server initialization timeouts

Qwen Code 0.21.8

Qwen Code 0.21.8 restores fork PR autofix, enables compression cache sharing across OpenAI/Gemini/Vertex, and improves /review performance to 93.3% cache hit rates

llama.cpp b10329

Corrects server info endpoint to report the isolate working directory when a tools runtime is configured

openclaw 2026.6.34

Hardens browser and network boundaries: sandboxed routes, trusted DNS, custom origins, and loopback endpoints now reject unsafe paths

ComfyUI 0.31.0

Reduces pinned memory on Linux systems without swap partitions

Claude Agent Sdk Typescript 0.3.225

Fixed background subagents failing to resume after shell commands or monitors complete in headless SDK sessions

Claude Code 2.1.225

Claude Code 2.1.225 fixes OAuth token handling in headless sessions, MCP keychain timeouts, safety-filter retry loops, and Remote Control session resume breakage on large conversations

Qwen Code 0.21.7-nightly.20260808.4ec0371e6

Qwen Code fixes core stability issues including file cache bounds, session transcript handling, memory reporting, and adds compression cache sharing with OpenAI and Gemini providers

llama.cpp b10326

TTS timing now accounts for full vocoder pass instead of single trailing window

Anthropic Sdk Csharp 12.40.0

C# SDK adds mid-conversation tool changes, session budgets, advisor tools, and GitHub skill auto-loading for Claude API integration

Anthropic Sdk Typescript 0.116.0

Adds session budgets, mid-conversation tool changes beta, advisor tool, and pinned inference location

LocalAI 4.8.2

LocalAI 4.8.2 adds NVIDIA NeMo-Speech.cpp backend for speech processing and improves model gallery resilience with mirror fallbacks

langchain-openai== 1.4.2

Filters langchain-generated content-block IDs; Handles ContextWindowExceededError

Cline Desktop 0.0.10

Cline Desktop adds OAuth authentication for remote MCP servers, fixes silent failures in chat turns, improves model onboarding visibility, and stabilizes reasoning controls across AI SDK providers

opencode 1.18.15

Message chronology fixes prevent ordering corruption during imports, forks, and compaction; Blob attachments now work in web UI; Broader locale support added to desktop

Anthropic Sdk Go 1.62.0

Adds session budgets, advisor tool, mid-conversation tool changes beta, and pinned inference location

Claude Code 2.1.224

Claude Code adds self-hosted environments, cross-session messaging, archive plugin installation, and sandbox credential-masking for Team/Enterprise users

Qwen Code 0.21.7

Removed 50-turn limit for Goals, added inline terminal image rendering, introduced repository-specific review manifests, and fixed streaming response hangs

LocalAI 4.8.1

LocalAI 4.8.1 fixes VRAM handling for malformed GGUF metadata, restores vllm-cpp backend ABI compatibility, and adds Qwen3.5 variants to model gallery

Kimi Code 0.34.0

Kimi Code adds Windows Computer Use support, persistent failure cards with one-click resume, and cache-expiry reminders for idle sessions

Cline 4.1.6

Cline adds Muse Spark 1.2 model option and fixes telemetry attribution to report failures against actual models used

Cline CLI 3.0.51

Reasoning effort now applies consistently across all providers including Ollama, new Muse Spark model added to Cline provider, error telemetry tracks actual model used

Qwen Code Live Host 0.1.0

Qwen Code Live Host 0.1.0 adds local GitHub authentication, mid-turn read-only command execution, and dependency closure testing for code reviews

llama.cpp b10290

AMD ROCm CI support for gfx1151; No inference changes

Qwen Code 0.21.6-nightly.20260806.cb3dc107f

Upstream release notes did not include a readable summary.

Cline 4.1.5

Cline 4.1.5 unifies reasoning settings across AI providers and adds clarity when free model promotions expire, replacing generic errors with model-selection prompts

llama.cpp b10289

Fixed directory traversal vulnerabilities in file search, improved path handling on Windows with UTF-8 conversion, and added cache expiration for UI picker searches

langchain-anthropic== 1.5.4

Langchain-anthropic 1.5.4 fixes tool schema composition handling, preserves tool_choice settings, and adds user_profile_id attribute support

Mistral Vibe 2.24.0

Adds admin config layer and server-side default model routing; Unpinned default model option tracks recommended model

HelixML 2.12.2

Autonomous bot runs on internal repos, recurring tasks work with external agents, org-level agent provider isolation, and UI refinements for task status and browser navigation

Elevenlabs Android 0.12.0

Adds media audio stream option to ConversationConfig; Fixes WebSocket disconnect attribution

LocalAI 4.8.0

LocalAI 4.8.0 ships vllm.cpp (C++20 inference engine), 3D generation, audio.cpp multi-family backend, VRAM budgets, and distributed mode hardening across 386 merged PRs

Cline Desktop 0.0.9

Universal macOS binary, session filtering, context-window cost ring, skill/workflow disambiguation, plugin naming, plan-mode file-blocking, auto-context recovery, and multi-provider empty-response retry

Kimi Code 0.33.0

CLI now runs on agent-core-v2 engine, adds Computer Use and WebBridge to marketplace, and improves session forking and trust prompts

Qwen Code 0.21.6-preview.0

Browser extension diagnostics, headless Goal workflows, PR/issue URL artifact tracking, web-shell agent stability fixes, and prompt cache reuse improvements

Claude Code Action 1.0.185

Claude Code Action now derives trigger timestamps for events, pins bun config for MCP servers, matches downloaded images ID, and validates collaborator permissions for workflow runs

Claude Agent Sdk Typescript 0.3.222

Fixed resumed agent sessions not inheriting user settings like API keys, environment variables, hooks, and permissions

Room Connect Ecr 0.5.0

Point-cloud generation for 3DGS initialization with curvature-weighted PLY, k-NN density stats, and Fast/Standard/Dense presets; Adds equirectangular backdrop image

llama.cpp b10271

Adds a per-conversation working directory with file picker; Agents can now treat path-like queries as directory navigation

Ollama 0.32.6

Ollama 0.32.6 speeds up Qwen3.5 on Apple GPUs via speculative decoding, fixes OpenAI API streaming format compliance, and improves TUI usability

HelixML 2.12.1

Chat UX improvements and rich attachments

llm CLI 0.32

Major overhaul of how prompts, responses, tools, and logs are represented; Backwards-compatible

whisper.cpp 1.9.2

Syncs with ggml; Includes Parakeet in build-xcframework.sh and fixes pkgconfig configuration

Elevenlabs Mcp 0.12.2

@cwood-11 made their first contribution in 128

Kimi Code 0.32.0

Kimi Code 0.32.0 adds hook events for session lifecycle tracking, offline model catalog fallback, renames loop control config keys, and fixes tool call ID parsing with colon-containing provider responses

Qwen Code 0.21.5

Adds tool-call outcome tracking, prevents unsafe MCP replays on connection loss, stops infinite Goal retries, and improves review result visibility

openclaw 2026.7.1-1

Openclaw fixes GPT/Codex response completion, memory startup recovery, WSL permissions handling, legacy migration tolerance, and plugin update staling

Claude Code 2.1.221

Claude Code 2.1.221 adds Focus view for cleaner chat UI, sandbox credential masking on Linux/WSL, permission-check security fixes for Bash and PowerShell, and MCP server connection reliability improvements

Claude Agent Sdk Typescript 0.3.221

Improved skills validation to reject malformed names, fixed external MCP servers not connecting before first turn preventing tool calls

Room Connect Ecr 0.4.0

Adds OPENCV camera model, HDR EXR output, and persistent camera/light settings for Gaussian Splat workflows

Hermes Agent 0.20.0

V0.20.0 Herald release; 3,650 commits, 559k insertions since v0.19.0

Mistral Vibe 2.23.3

/retry command and improved /rewind control surface let users recover from interrupted LLM responses, with better cost visibility and cache token accounting

llamafile 0.10.5

Focused on simplifying the upstream llama.cpp sync task

llama.cpp b10236

Implements DSv4 Lightning Indexer and f16 Lightning Indexer for Metal; Affects 128-dimensional 64-head inputs on Apple silicon

Qwen Code 0.21.4

Web Shell is now a native desktop app with lifecycle management and auto-updates, subagents are isolated per fork, and prompt cache reuse reduces latency

llama.cpp b10232

WebGPU backend adds f16 repeat support; MacOS KleidiAI build disabled

openclaw 2026.7.2-beta.7

OpenClaw adds durable message recovery across gateway restarts, conversation branching and rewinding, MCP app hosting on dashboards, agent approvals with push notifications, Teams/Zoom meeting integration, Wear OS companion app, and local LLM inference

Cline CLI 3.0.49

Cline CLI 3.0.49 fixes undo/checkpoint functionality, improves Ollama cold-load handling, adds UTF-8 PowerShell stdin support, and resolves provider error reporting

Cline SDK 0.0.69

Ollama timeout increased to 5 minutes for large model cold-loads, empty responses retried automatically, checkpoints now captured and restored with file state tracking

Qwen Code 0.21.3-nightly.20260802.184365390

Fixed history pagination failure on oversized transcripts and completed TUI keyboard shortcut documentation

NanoClaw 2.1.54

Rollup of 36 commits since v2.1.17; No individual fix highlighted

llama.cpp b10219

Llama.cpp now preserves reasoning content in chat history, enabling multi-turn conversations to leverage previous model reasoning with --reasoning-preserve flag

opencode 1.18.11

Fixed MCP SSE reconnect loops, reasoning field parsing, and desktop UI issues including external link handling and file tree clipping

koboldcpp 1.118.1

Row Split option removed; Selecting it now defaults to tensor split

Cli 1.0.19

Adds Hailuo H3 video generation; Fixes quota reporting and truncated downloads

openclaw 2026.7.2-beta.6

OpenClaw ships crash-recovery for persisted data, durable message delivery across channels, conversation branching, interactive MCP dashboards, and local inference support with Claude Opus 5

llama.cpp b10212

Adds driver-version check for Intel GPUs on Windows to avoid crashes; No action unless running that config

InvokeAI 6.14.0

Adds video generation via Wan 2.2

Kimi Code 0.31.1

Fixes model catalog initialization race, preserves interrupted outputs, enables Monaco code highlighting, improves TUI performance, and corrects permission ordering and chat rendering bugs

ComfyUI 0.29.2

Frontend improvements and new API/partner nodes for extending ComfyUI's extensibility and integration capabilities

Qwen Code 0.21.2

Autofix now defers lower-severity suggestions after five rounds, Auto Recall searches external context with security limits, and Web Shell gains contextual task panels

Maple 3.3.1

Maple 3.3.1 adds private Voxtral text-to-speech, improves Agent Mode continuity, hardens Local Proxy authentication, and restores Android 16 KB page-size support

Cline CLI 3.0.48

History command now opens in existing TUI; Connector threads recover from missing sessions

Cline SDK 0.0.67

Reasoning controls now driven catalog with normalized budgets per provider, MCP timeouts configurable per server, session forking for editing prior prompts, and checkpoint compaction fixed to prevent repeated summarizer calls

Qwen Code 0.21.1-nightly.20260731.702932cc7

Qwen Code adds session-scoped MCP runtime, persists Goal v3 state, fixes shell output flushing, and improves CI reliability with zombie process cleanup and retry logic

Hermes Agent 0.19.1

Hermes Agent 0.19.1 stabilizes 1,000+ PRs with bug fixes across gateway, voice, desktop, and installer subsystems

llama.cpp b10195

Tests alternative convolution layout; Extends layout checks for conv2d kernel

llm CLI 0.32rc1

Default model now GPT-5.6 Luna for users without a configured default

Elevenlabs Mcp 0.11.0

Adds simulate conversation tool, defaults STT to scribe v2

Elevenlabs Python 2.60.0

Fixes blocking in AsyncConversation startup, corrects multipart field handling

Mistral Vibe 2.23.2

Adds skill-creator built-in flow, session cache-hit token tracking, custom login domain support

Kimi Code 0.31.0

Kimi Code adds plugin-contributed custom agents with Markdown definitions, configurable system prompts, and sub-agent delegation, plus fixes for request headers and task blocking

Qwen Code 0.21.1-nightly.20260730.1643a6c9a

Qwen Code adds session-scoped MCP runtime, Goal v3 state persistence, image zoom tooling, and fixes shell stream flushing, quota handling, and composer text preservation

llama.cpp b10182

Adds NextN and MTP speculative decoding for GLM-5.2; GLM DSA support

Kimi Code 0.30.0

Customizable footer status line, halted retries on invalid tool calls, quota warnings for plugins, fast-fail on account exhaustion, and removed file upload size limits

Cline Desktop 0.0.7

Cline Desktop adds system tray status, paginated history, session favorites, subagent tracking, free-tier model support, fixes regional endpoint routing and tool-call parsing bugs

ComfyUI 0.29.0

Streams video transcode instead of buffering all frames in memory

Qwen Code 0.21.1

Qwen Code adds Goal v3 runtime orchestration, GitHub channel notifications, workspace trust hot-reloading, and read-only external corpora search for code review automation

Cline 4.0.12

Cline adds free model tier support with usage limit indicators and fixes response loss during max-turns errors

llama.cpp b10173

Added Laguna-S-2.1 model support to llama.cpp for local inference

opencode 1.18.8

MCP server reconnection after SDK expiration, OAuth callback port handling, and deprecated sampling parameter removal improve compatibility and stability across newer AI model integrations

Dify 1.16.1

Dify 1.16.1 adds workflow tool multi-select inputs, node locators in logs, agent DSL export, and fixes collaboration draft saves, RAG document processing, and agent preview chat persistence

openclaw 2026.7.2-beta.5

OpenClaw 2026.7.2 adds crash-durable message recovery, conversation branching, interactive MCP dashboards, structured agent approvals across platforms, and local model inference with Wear OS support

Cline Desktop 0.0.6

Cline Desktop adds queued message management, persistent update notifications with one-click restart, and improved UI layout for narrow windows and dark mode

Packages React 1.11.0

Streaming reasoning response events now available via onAgentReasoningResponsePart callback

Qwen Code DSW manual benchmark POC 20260727-2

Nightly achieved 376/500 resolved on SWE-bench Verified, results quarantined pending production validation

Room Connect Ecr 0.3.0

Depth map rendering 32x faster. Hot-swap GLB workflow with confirmation dialog

llama.cpp b10153

MiMo-V2.5 RVQ-based audio input support added

Kimi Code 0.29.2

Kimi Code 0.29.2 fixes goal interruption, message handling during active goals, chat copying over HTTP, and /undo consistency; Defers tool schema registration and improves runtime state management

vLLM 0.26.0

VLLM 0.26.0 adds Inkling model support, DeepSeek-V4 performance optimizations, fp32 generation heads, flexible attention backends, and mature KV offloading with tiered storage

Qwen Code 0.21.0-nightly.20260727.c003e1718

Measure insight days and hours in local time everywhere; Extract review verification runner; Don't fail triage cleanup when there is nothing to clean; Align GenAI content telemetry fields

AI release Jul 26, 2026

Maple 3.3.0

Adds local PDF OCR with PDFOxide, expands Agent Mode with authenticated web tools, stabilizes chat-history pagination

AI release Jul 26, 2026

Qwen Code 0.21.0

Qwen Code 0.21.0 adds workspace management, file previews, background agent persistence, MCP reconnects, and autofix improvements for stale bases and infrastructure failures

AI release Jul 25, 2026

vLLM 0.26.0

411 commits from 212; Check release notes for model and hardware support changes

AI release Jul 25, 2026

Ollama 0.32.4

Ollama 0.32.4 adds Laguna support for Apple GPUs, fixes Qwen3 MoE expert quantization, and improves speculative decoding performance

AI release Jul 25, 2026

Qwen Code 0.21.0

Qwen Code 0.21.0 adds workspace management, file previews, background agent persistence, MCP reconnects, and auto-recovery for stale pull request bases

opencode 1.18.5

Opencode 1.18.5 stabilizes multi-model AI handling (Claude, OpenAI, Mistral, MiniMax) and adds current-server support to the desktop app with backwards compatibility

langchain-anthropic== 1.5.2

LangChain Anthropic added Claude Opus 5 model support for AI tooling integrations

Cline 4.0.11

Adds Claude Opus 5 support across providers, including 1M context variants

Claude Code 2.1.219

Claude Opus 5 now default with 1M context, sandbox network allowlist controls added, nested subagent forwarding in stream-json, and various fixes for permission handling and UI state

Claude Agent Sdk Typescript 0.3.219

Adds interrupt cancellation, fast-mode diagnostics, directory lifecycle hooks, network allowlist controls, and workflow sizing guidance for agent orchestration

Anthropic Sdk Csharp 12.39.0

C# SDK adds Claude Opus 5 model, tool addition/removal blocks, tool_change events, and fixes streaming parallel tool-use logic

HelixML 2.11.58

HelixML adds Slack topic delivery, fixes clipboard paste in embedded streaming, and defaults to Claude Opus 1M context

llama.cpp b10105

Refactors mlock, mmap, and direct I/O into unified load-mode argument

Qwen Code DSW SWE-bench Full POC 20260724 Run 3

Isolated test pre-release for SWE-bench PR 7656; Not a production artifact

Kimi Code 0.29.1

Kimi Code 0.29.1 adds MCP server timeouts, fixes reasoning field detection for OpenAI-compatible endpoints, and enables web search/fetch configuration via environment variables

Qwen Code 0.20.1-nightly.20260724.7d17c44a3

Qwen Code nightly adds workspace switching, lazy-loads dependencies for faster startup, optimizes session rendering, and includes security fixes for daemon secret isolation

llama.cpp b10099

CUDA NVFP4 W4A4 activation quantization improvements, fused per-channel amax and quantization kernels

langchain-openai== 1.4.1

LangChain OpenAI adds LangSmith gateway support via environment variable and fixes GPT-5.3 profile configuration

langchain-anthropic== 1.5.1

LangChain's Anthropic integration now supports LangSmith gateway via environment variable and enables structured output for Claude Opus 4.8

Mistral Common 1.11.7

Mistral Common adds auto-detecting chat template generation, validates special token policies during decoding, and rejects thinking chunks in tool messages

HelixML 2.11.57

API thread hang fixes, GitLab topic webhooks, Claude subscription clarification, and 100MB attachment support for spec tasks

Jan 0.8.4

Settings and credentials now live in backend-managed store. Migration required

Qwen Code 0.20.0-nightly.20260722.b98306b7e

Qwen Code benchmark; Field; Result; Status

Ollama 0.32.3

Ollama 0.32.3 fixes stalled downloads, expands GPU support to Windows ARM64 and B200, and adds chat/tool-calling for Laguna 2.1 models

Claude Agent Sdk Python 0.2.126

SDK now surfaces query termination reasons and properly types model usage with provider identification for cross-provider model tracking

Claude Code Action 1.0.181

Claude Code Action now shares WIF credentials across spawned processes instead of creating duplicates

llama.cpp b10090

Adds depthwise conv2d kernel for WebGPU backend, ported from Vulkan

Anthropic Cli 1.19.0

Adds support for Managed Agents model effort, initial session events, threads delta streaming

Anthropic Sdk Csharp 12.37.0

C# SDK adds Managed Agents model effort controls, initial session events, and threads delta streaming support

Anthropic Sdk Typescript 0.113.0

Adds support for Managed Agents model effort, initial session events, threads delta streaming

Kimi Code 0.29.0

Kimi Code 0.29.0 adds granular thinking effort levels in ACP clients, custom Markdown-based agents with tool gating, video attachment support, and raw file serving via API

Elevenlabs Python 2.59.0

Adds HMAC-signed post-call webhook config for on-prem conversations

Maple 3.2.0

Adds manual Agent project ordering and prevents Markdown image loading

Qwen Code 0.20.0-preview.0

Upstream release notes did not include a readable summary.

Cline Desktop 0.0.2

First signed, notarized macOS build (Apple Silicon and Intel) for running and inspecting Cline sessions

InvokeAI 6.13.7

Closes config-file exposure affecting all versions through 6.13.6. Upgrade immediately

llama.cpp b10078

Vulkan queue refactoring enables driver-side synchronization on supported hardware, reducing host-side mutex contention for GPU inference

langchain-anthropic== 1.5.0

Adds reasoning-effort as a standard parameter; Fixes builtin tool recognition

langchain-openai== 1.4.0

Adds reasoning-effort parameter; Bumps Pillow to 12.3.0

Mistral Vibe 2.22.0

Adds /new alias for /clear and ACP review support with live refresh

HelixML 2.11.56

HelixML fixes task merge detection bugs, adds GitLab OAuth attribution, and requires branch specification for existing mode

Introducing Gemini 3.6 Flash 3.5

Google released Gemini 3.6 Flash with faster inference and new specialized variants (Flash-Lite, Flash Cyber) for different compute constraints

langchain-core== 1.5.0

Adds reasoning effort as a standard chat-model parameter

Qwen Code 0.20.0-nightly.20260721.cda0e0348

Qwen Code improves autofix label-driven workflows, adds persisted conversation branches, routes shell commands level, and isolates parallel web-shell sessions

Ollama 0.32.2

Ollama 0.32.2 adds agent skills system, fixes Anthropic thinking-block handling, updates llama.cpp and MLX backends, and enables unlimited tool rounds for cloud models

Hermes Agent 0.19.0

Hermes Agent drops first-token latency 80% across platforms, streams reasoning live, adds terminal subscription management, integrates Bitwarden/1Password, and ensures durable delivery through crashes

opencode 1.18.4

Adaptive thinking now works with Kimi models on Anthropic-compatible providers

Kimi Code 0.28.1

ACP sessions now accept pre-configured model credentials, web servers run foreground-only, and subagents respect runtime permission changes

Kimi Code 0.28.0

Replaced kimi server with kimi web for foreground operation and multi-instance support sharing one home directory, fixed permission mode descriptions, and enabled symlink handling in agent file loading

HelixML 2.11.54

Fixes macOS VM pinning, recovers idle threads, returns admin status in login, rejects nested YAML configs, and exposes external agent sessions in API

Qwen Code 0.20.0

Qwen Code 0.20.0 adds daemon log rotation, WebShell history replay, improved review findings tracking, and fixes Java SDK timeout handling plus tool calling schema support

Cline SDK 0.0.65

Claude and Codex provider SDKs now load on demand instead of bundling, reducing install size, with OAuth retry, team-run isolation, and Windows shell fixes

Qwen Code 0.19.12-nightly.20260719.86ad532de

Qwen Code adds daemon log rotation, fixes Java SDK timeout constants and exception handling, improves code review CI checks and findings tracking, and defers TUI startup for faster daemon initialization

openclaw 2026.7.1

Control UI overhaul, app updates across iOS/Android/macOS, GPT-5.6 and Tencent Hy3 support, improved coding agents, and fixes to Gateway crashes, messaging channels, and scheduled workflows

openclaw 2026.7.2-beta.3

Remote coding sessions on cloud workers, native mobile automations, safer channel operation, guided setup flows, and gateway recovery improvements

Cli 0.5.6

Agent templates fixed to work correctly with CLI tooling

llama.cpp b10067

K/V cache rotation fix for DFlash quantization workflows

Open Interpreter 0.0.34

Open Interpreter 0.0.34 release

Qwen Code 0.19.11-nightly.20260718.767a32484

Qwen Code adds multi-workspace hardening, session export, skill management UI, and improves daemon startup tracing and stream handling

Open Interpreter 0.0.30

Open Interpreter 0.0.30 release

Buffa 0.9.0

Changie fragment files adopted to end changelog merge conflicts. Codegen and struct evolution policy documented

Mistral Vibe 2.21.0

Breaking change: hooks graduated from experimental, hooks.toml is now stable. Registry skills content store and ACP config schema added

llama.cpp b10064

OpenCL kernel gemm_moe_q6_k_f32_ns now loaded from binary kernel library. Binaries for macOS, iOS, Linux, and Windows published

Mistral Common 1.11.6

Fixes audio chunk validators, rejects invalid sampling rates, improves token decoding under strict policies

Dify 1.16.0

Dify 1.16.0 ships Agent sandbox execution with skill system, workflow integration, and improved AI-generated workflow generation for faster agent building

Open Interpreter 0.0.29

Open Interpreter 0.0.29 release

openclaw 2026.7.1

OpenClaw 2026.7.1 overhauls UI and onboarding, adds GPT-5.6 and Tencent Hy3 support, improves Codex agents and messaging channels, and fixes Gateway crashes

Open Interpreter 0.0.28

Open Interpreter 0.0.28 release

Elevenlabs Android 0.11.1

Fixes R8/ProGuard obfuscation of outgoing event fields

Open Interpreter 0.0.27

Open Interpreter 0.0.27 release

Cline CLI 3.0.44

Cline CLI 3.0.44 fixes UTF-8 BOM parsing in config files and improves token limit handling across LLM providers

Cline SDK 0.0.64

Cline SDK improves token limit handling across AI providers and fixes UTF-8 BOM parsing in config files

Qwen Code 0.19.11

Qwen Code added extension management, workspace locking, shell heartbeats, structured memory, and PR preview automation for improved development workflows

Claude Agent Sdk Typescript 0.3.212

Fixed CLI argument parsing for session flags and added resolved model visibility in multi-model agent tool outputs

Claude Code 2.1.212

/fork separates conversation branches into background sessions, while /subtask replaces in-session subagents; Adds runaway-loop safeguards, long-running tool backgrounding, and session-resumption picker

llama.cpp b10051

MacOS, iOS, and Linux build artifacts now included in release notes

llamafile 0.10.4

Introduces first version of transcribefile

Anthropic Sdk Java 2.49.0

Anthropic Java SDK adds dreaming and MCP Tunnels support, plus session delta accumulator for agent message previews

Anthropic Sdk Php 0.37.0

Anthropic PHP SDK adds dreaming and MCP Tunnels support, plus session delta accumulator for agent message previews

Cli 1.0.18

Add npm package metadata; .github/workflows: Migrate workflows to Blacksmith runners 203; Release: v1.0.18

Kimi Cli 1.49.0

Kimi CLI 1.49.0 fixes context budget allocation, reasoning content handling, and aligns telemetry events with schema for better observability

Hugging Face Transformers Patch release 5.14.1

Fixes assisted generation issue with EncoderDecoderCache models

Ollama 0.32.1

Ollama 0.32.1 fixes MLX model cache leaks, improves Gemma 4 tool calling, and adds working directory context to agents

Cline CLI 3.0.42

Ollama native API routing fixed to properly respect context window and timeout settings

Kimi Code 0.25.0

Chat now accepts any file type, uploading unconsumed files to server for model access, plus fixes for authentication bypass, symlink traversal, Anthropic thinking controls, and CSP issues

Qwen Code 0.19.10-nightly.20260716.506ce0a1a

Web-shell improvements including workspace locking, extension management, adaptive composer layout, plus core heartbeat liveness detection and structured channel memory management

llama.cpp b10032

Fixes OpenCL buffer creation for CL 2.x; MacOS arm64 and Linux CPU builds included

Maple 3.1.5

Adds MCP server support; Hardens Zapstore zsp install

Hugging Face Transformers 5.14.0

Adds Inkling multimodal model: 975B total parameters, 41B active, accepts text, image, and audio inputs

Cline SDK 0.0.61

Context compaction now reports progress; Workspace git state persisted across sessions

ComfyUI 0.28.0

Fixes Qwen3-VL tokenizer crash with custom embeddings; Adds AGENTS.md documentation

Mistral Vibe 2.20.0

Windows command prompt now shell-aware; Double-Esc triggers rewind when input is empty; Hardened bash tool permissions

Qwen Code 0.19.10

Multi-workspace support extends across transports and sessions, Web Shell gains artifact panels and user-level model config, and session recovery becomes more resilient to failures

Claude Code Action 1.0.174

Claude Code Action now redacts GitHub user-to-server tokens from logs to prevent credential exposure

Claude Agent Sdk Typescript 0.3.210

Bash tool output now includes timedOutAfterMs field when commands are auto-backgrounded due to timeout

llama.cpp b10012

Adds contiguity checks for inner tensor dimensions

Room Connect Ecr 0.1.2

Add K8s testing POR and upload optimization POR ( )

LocalAI 4.7.1

Scopes llama.cpp serving options to the llama.cpp code path only

Room Connect Ecr 0.1.0

Switched to GitHub-hosted runners with OIDC support to unblock ECR image pushes that were queuing indefinitely on self-hosted infrastructure

HelixML 2.11.51

HelixML adds reasoning effort controls, subscription usage tracking, cron message configuration, and improved org chart visibility for agent workflows

Kimi Code 0.24.1

Kimi Code fixes session hangs from empty reasoning steps, corrects tool availability timing, improves thinking-level persistence and UI display across web and CLI

vLLM 0.25.1

VLLM 0.25.1 fixes FFmpeg import blocking model startup and mixed-dtype allreduce corruption in quantized models

Kimi Code 0.24.0

Kimi Code 0.24.0 defaults local server to agent-core-v2 engine, handles long-running commands via backgrounding instead of killing, and adds session diagnostic export

Qwen Code 0.19.10

Multi-workspace support now spans transport and sessions, Web Shell gains artifact panels and editable model config, and session recovery is more resilient to stream failures and daemon restarts

Claude Code 2.1.209

Fixed dialog blocking in Claude agents background sessions that prevented /model and other commands from executing

openclaw 2026.7.1

OpenClaw 2026.7.1 overhauls Control UI and onboarding, expands model support including GPT-5.6 and Tencent Hy3, strengthens coding-agent workflows, and improves Telegram, Slack, Discord, and Apple Messages integrations

llama.cpp b9992

Corrects server endpoint to honour per-request reasoning-budget tokens in chat completions

Elevenlabs Python 2.58.0

Upstream release notes did not include a readable summary.

openclaw 2026.7.1-beta.6

Adds Claude Sonnet 5, Meta Muse Spark, and new providers; Improves mobile offline chat, session management, macOS app, and startup recovery

AI release Jul 12, 2026

llama.cpp b9977

OpenCL int8 dp4 dense and MoE GEMM optimization for Adreno GPUs

AI release Jul 12, 2026

Kimi Code 0.23.6

Dismissed prompts now treated as non-answers, print-mode sessions survive background tasks and active goals, subagent timeouts extended to 2 hours and made configurable, web UI connection and scrolling fixes

Maple Proxy 0.1.11

Maple Proxy v0.1.11; All binaries include

vLLM 0.25.0

558 commits, 232; Compatibility and performance sweep

OpenClaw 2026.7.1-beta.5

Conversational AI onboarding, multi-provider routing with quotas, offline mobile chat with speech, session organization with auto-titles, and Telegram/Slack integration across platforms

Hugging Face Transformers Patch 5.13.1

Compatibility patch for vLLM 0.25.0

llama.cpp b9959

Llama.cpp synced ggml changes, updating core ML infrastructure for model inference

Claude Agent Sdk Python 0.2.116

Bundles Claude CLI 2.1.207; Fixes CI workspace trust for project-scoped permission grants

Ollama 0.32.0

Adds Qwen3.5 parser and renderer selection; Agent UI warns before loading old agent models

Qwen Code 0.19.9

Fixes subagent tool-call loops, broken session history detection, vision bridge timeouts, and shell process management while adding configurable memory agent timeouts

langchain== 1.3.13

LangChain adds Meta model support via new meta extra and implements explicit prompt caching for OpenAI models

Qwen Code 0.19.9

Fixes subagent tool-call loops, broken session history detection, vision bridge timeouts, and shell process management; Adds configurable memory agent timeouts and session persistence

llama.cpp b9952

Llama.cpp optimizes DeepSeek V4 inference with f16 KQ masks, removes unnecessary attention bias, and eliminates redundant cache repeats for faster execution

Qwen Code PR #6633 Screenshots

Screenshots for PR 6633 verification

Claude Code 2.1.206

Claude Code adds directory suggestions to /cd, auto-confirms git pushes to configured remotes, fixes MCP timeouts and OAuth re-authentication, and improves agents view usability

Claude Agent Sdk Typescript 0.3.206

Added command_lifecycle frames tracking message state transitions, eliminating stale duration metrics when API returns no results

Qwen Code 0.19.8-nightly.20260710.205430235

Fixes subagent tool-call loops, broken history chain detection, vision bridge timeouts, and CLI command matching for more reliable agent execution

llama.cpp b9945

Hexagon: tiles wide rows in unary ops to avoid VTCM overflow, replaces divs with fastdiv, adds VTCM layout and host-computed offsets

Kimi Code 0.23.4

Kimi Code adds fuel-pack balance display, fixes session resumption and prompt-goal execution, implements image compression with configurable size limits to prevent request failures, and improves web UI reliability and styling

HelixML 2.11.47

HelixML fixes Anthropic model ID rewriting for Zed Agent, surfaces deployment logs, and prevents public 502 error leaks

llm CLI 0.31.1

Fixes JSON error from OpenAI Chat Completion providers when tool calls have empty arguments

koboldcpp 1.117.1

KoboldCpp 1.117.1 fixes thinking trace display, improves multimodal image placement, adds full Ollama compatibility with embeddings and tool calling, and reworks group chat with multi-character avatars

koboldcpp 1.117

Fixes terminal output for thinking traces; Multimodal images now inserted inline instead of at context top

Dify 1.16.0-rc1

Experimental Dify Agent support; Serve only to trusted users

Mistral Vibe 2.19.1

Bash tool for shell sessions, per-tool description overrides, reactive compaction, and lazy history loading improve agentic workflows and startup performance

Elevenlabs Js 2.57.0

Pins Node to lts/Krypton in CI; Publishes with bundled npm

Cline CLI 3.0.39

Cline CLI 3.0.39 adds free model selection in ClinePass provider, fixes str_replace diffs and context compaction, and improves telemetry tracking with versioned client headers

langchain-openai== 1.3.4

Suppresses Pydantic serializer warnings on structured output parsed fields and fixes async context loop handling

langchain== 1.3.12

LangChain 1.3.12 fixes interrupt propagation in tool retry middleware, prevents process-group kills in shell operations, and sanitizes Anthropic cache markers on fallback retries

Qwen Code 0.19.8

CLI now supports inline skill preview with editor handoff, serve environment isolation with admission control, and multi-workspace session routing

llama.cpp b9910

Fuses MMVQ post-scale and batch-size-1 path for NVFP4

Hermes Agent 0.18.2

Fixed WhatsApp bridge dependency resolution Baileys from git commit to published npm release, enabling reliable Docker builds

Qwen Code 0.19.7

Qwen Code 0.19.7 strengthens PR triage, adds session management and export, fixes file restoration during autofix, and improves daemon status visibility

MLX 0.32.0

Minor bump to 0.32.0; Qmm implementations now generated via cmake

Claude Code 2.1.203

Claude Code 2.1.203 fixes session recovery, background agent stability, and context-usage performance regressions affecting multi-agent workflows and daemon reliability

Claude Agent Sdk Typescript 0.3.203

SDK now emits background_tasks_changed messages for full task state tracking and fixes TypeScript type reference resolution in stable builds

Vla Evaluation Harness 0.4.0

Recording on; LeRobot bridge ships with four reproduced LIBERO scores and reproducible Docker builds

HelixML 2.11.46

HelixML adds LM Studio integration with autodetection, improves sandbox stability, and unifies agent messaging on session-scoped queues

Kimi Code 0.23.1

Kimi Code 0.23.1 fixes goal token budgets, WebSocket reconnection, background subagent handling, and Claude thinking preservation on Anthropic providers

Cline CLI 3.0.38

Cline CLI adds visual mode distinction (blue for act, amber for plan) across UI, improves input parsing tolerance for weaker models, and defaults unspecified models to 128K context

Qwen Code 0.19.6-nightly.20260707.bcdb44c5d

Qwen Code tightens PR review gates with improved problem detection, adds daemon status monitoring and session management, fixes auth persistence and file restoration bugs, and optimizes glob traversal performance

Claude Code Action 1.0.166

Fixes symlink handling in config snapshots, preserves repeated CLI flags, and improves error propagation in install pipeline

Ollama 0.31.2

Flash attention now works on older NVIDIA GPUs, vision model offloading improved, and non-UTF-8 model paths fixed

llama.cpp b9890

Adds Metal col2im 1d op support for f32, f16, bf16

LocalAI 4.6.2

Use slices.Contains to simplify code; Shard single-arch backend ma

Cli 0.5.5

Fixes agent create and update to send raw JSON, preserving config round-trip

Kimi Code 0.23.0

Preserved thinking enabled, progressive tool disclosure for prompt caching, archived session management, and improved user question feedback in AskUserQuestion

Client Python 2.6.0

Adds PII and secrets redaction when OpenTelemetry is active

HelixML 2.11.45

HelixML fixed node discovery query performance and updated error message handling in wedge-breaker tests

Qwen Code 0.19.6-nightly.20260706.47f62a466

Qwen Code improves PR review triage, adds daemon status monitoring and session management, fixes file-restore and auth bugs, and optimizes glob traversal performance

langchain-mistralai== 1.1.6

Surfaces citation metadata from chat responses. Model profile data refreshed

langchain-openrouter== 0.2.6

Adds support for custom HTTP header injection via default headers

InvokeAI 6.13.6

Fixes dictionary key error crash in Qwen Image and Qwen Image Edit models with default single-file encoder

llama.cpp b9878

Fixes stale tensor-split metadata for GQA attention on draft models

openclaw 2026.7.1-beta.2

OpenClaw adds GPT-5.6 support, external harness attachment for resumable workflows, Telegram Codex integration, event-driven cron scheduling, and macOS local Gateway setup

InvokeAI 6.13.5

InvokeAI 6.13.5 stabilizes image generation with bug fixes for prompt history, GPU crashes, and UI layout while updating React 19, ROCm 7.1, and Transformers 5.5.4

llama.cpp b9873

Adds guard to prevent dereferencing unallocated K/V rotation buffers; Defensive fix for edge-case crashes

Checkpoint Engine 0.4.2

ParameterServer metadata now exposed via HTTP, enabling cross-process parameter sharing in distributed AI training setups

LocalAI 4.6.0

AMD ROCm backends now offload to GPU correctly, distributed worker failures no longer lock model loads, and realtime sessions warm pipelines eagerly to eliminate cold-start stalls

Cline CLI 3.0.37

Cline CLI now handles malformed tool calls from weaker models, makes mode switches visible to the model, and fixes plan/act mode prompt handling

Qwen Code 0.19.6

Qwen Code 0.19.6 adds nested sub-agents, dataviz skill, mobile improvements, tool execution timeouts, and fixes multimodal payload bloat

Claude Agent Sdk Typescript 0.3.201

Claude Agent SDK TypeScript now matches Claude Code v2.1.201 capabilities for building AI agents

Claude Agent Sdk Typescript 0.3.200

SDK now validates model inputs, fixes permission-mode callbacks, and accepts 'manual' as permission alias

Hugging Face Transformers 5.13.0

Adds KimiK 2.5, 2.6, 2.7 architecture support

Mistral Vibe 2.19.0

Adds --worktree option for git worktree creation, raises Leanstral context limit, disables Sentry auto integrations

llama.cpp b9866

Trims messages sent to StepFun parser, fixes long reasoning loops

Diffusers 0.39.0

Ships Cosmos 3 pipelines for NVIDIA's unified world foundation model

Qwen Code 0.19.5

Qwen Code adds HTTPS/TLS support, unified reasoning effort control, and leader approval workflows while fixing CLI responsiveness and web UI session handling

AI release Jul 2, 2026

Cline 4.0.6

Generalized model capability warnings to apply more broadly across different AI models

AI release Jul 2, 2026

llama.cpp b9860

Adds llama_model_ftype_name function to expose quantization type as string

AI release Jul 2, 2026

openclaw 2026.7.1-beta.1

Adds GPT-5.6 model recognition across catalog and runtime paths

Claude Code Action 1.0.162

Agent-approval-check composite action added to Claude Code Action for gating AI-driven code changes behind manual review

Claude Code 2.1.198

Background agents now auto-commit, push, and open PRs on code completion; Extended thinking and model selection now propagate to subagents; Transient network errors retry instead of failing

Hermes Agent 0.18.0

Major release: 1,720 commits, 998 PRs merged, 949 issues closed, 370

HelixML 2.11.41

Fixed agent session restart to properly clear state and corrected DALL-E model filtering for non-OpenAI endpoints

Mistral Vibe 2.18.4

Mistral Vibe fixes session resume behavior, tool rendering, and config validation while improving diff display and auto-approve in lean mode

Elevenlabs Python 2.56.0

Adds disable-auth parameter to speech engine

llama.cpp b9857

Uses Hugging Face primary split as model path; Fixes issue 25181

Qwen Code 0.19.4

Qwen Code 0.19.4 adds voice dictation to desktop, resumable interrupted turns without synthetic messages, Chrome extension via daemon architecture, and multiplayer channel-resident agent foundation

Open WebUI 0.10.2

Streamed reasoning display, folder uploads preserving structure, memory system refinements, environment-based API configs, and security fixes

Qwen Code 0.19.3-nightly.20260701.a974594d7

Voice dictation in desktop app, configurable auto-compact thresholds, resumable interrupted turns, and Chrome extension via daemon-direct architecture ship in this nightly build

ComfyUI 0.27.0

Adds int8 convrot model support

Ollama 0.31.1

Gemma 4 nearly 90 percent faster on Apple Silicon via multi-token prediction

Cline 4.0.5

Cline adds Claude Sonnet 5 support across eight providers with updated model picker and recommendations

Anthropic Sdk Php 0.34.0

Adds managed-agent event delta streaming, agent overrides, reverse pagination, vault credential scoping, webhook events

Anthropic Cli 1.14.0

Adds managed-agent event delta streaming, agent overrides, reverse pagination, vault credential scoping, webhook events

Claude Code 2.1.197

Claude Sonnet 5 now default in Claude Code with 1M-token context window and reduced pricing through August

LocalAI 4.5.6

LocalAI adds native voice and face detection backends, improves distributed state management across replicas, and fixes GGUF context sizing and gallery matching

llama.cpp b9849

Registers attention normalization tensor for Qwen3; Fixes input assignment in layer loop

Maple 3.1.3

Maple 3.1.3 rejects ARM64EC Windows runtime DLLs and simplifies Authenticode verification for improved binary integrity validation

Checkpoint Engine 0.4.2-rc0

Exposes ParameterServer metadata via HTTP for cross-process join workflows

Vla Evaluation Harness 0.3.0

Adds SQLite recording and experiment tracking for observability workflows

InvokeAI 6.13.5

Maintenance release focused on bug fixes and stability

Qwen Code 0.19.3

Qwen Code 0.19.3 fixes streaming timeouts, web fetch fallbacks, IDE port validation, and skill persistence workflows while adding daemon voice APIs and skill usage stats

Claude Code 2.1.196

Organization admins can now set default models, sessions have readable names, file attachments are clickable, and numerous fixes improve security, reliability, and background job handling

vLLM 0.24.0

571 commits from 256; No breaking changes called out in summary

Open WebUI 0.10.0

Adds folder sharing with teams

Ollama 0.30.12

Fixed tool call parsing to handle JSON strings with braces correctly, and updated llama.cpp and MLX dependencies

Maple 3.1.2

Maple 3.1.2 fixes AppImage runtime helpers and establishes signed Windows release verification with hardened Authenticode validation

llama.cpp b9839

Adds DeepSeek V4 conversion support including pro model, rope fixes, and chat template

Kimi Code 0.20.2

Adds optional API parameter to exclude empty sessions from the session list

Cline CLI 3.0.32

Cline CLI 3.0.32 improves ClinePass onboarding with intermediate steps, refined UI copy, better context compaction, and clearer error messages

Cline 4.0.2

Cline 4.0.2 adds reasoning effort support for DeepSeek thinking models and fixes environment variable replacement in the webview

openclaw 2026.6.11-beta.2

Slack/Mattermost relay modes, per-DM model overrides, file-driven agent workflows, and reliability fixes for channel delivery, WhatsApp context preservation, and agent fallback behavior

llama.cpp b9835

Fixes UI stop and reasoning skip in single-model mode

Qwen Code 0.19.3

Qwen Code 0.19.3 adds workspace voice and control APIs, skill usage stats, fixes streaming timeout handling and reasoning content preservation, and improves web shell transcript and table interactions

Qwen Code 0.19.2-nightly.20260628.714513df2

Qwen Code 0.19.2 fixes web fetch JSON fallback, adds daemon voice/control APIs, implements skill confirmation before persistence, and improves streaming stability and CLI responsiveness

koboldcpp 1.116.1

Koboldcpp 1.116.1 adds Krea 2 Turbo, Ideogram 4, and Boogu Edit image generation models, MCP server support in the web UI, and fixes MTP performance issues

koboldcpp 1.116

Adds Krea 2 Turbo support

LocalAI 4.5.5

Fixed CI build failures across multiple backends, whisper macOS library loading, and added new gallery models

llama.cpp b9825

Vulkan backend fix for zero-input step operator handling in llama.cpp inference

Qwen Code 0.19.2-nightly.20260627.d93bec905

Qwen Code adds daemon voice and control APIs, skill usage stats, streaming timeout handling, web_fetch JSON fallback, and fixes CLI/IDE stability issues including duplicate responses and port validation

LocalAI 4.5.4

LocalAI 4.5.4 fixes Darwin backend binary derivation from exec lines, resolving macOS model execution issues

llama.cpp b9821

Adds support for version, licenses, and help flags

Claude Code Action GitHub 1.0

1.0 is generally available

Claude Code Action 1.0.159

Claude Code Action fixes app token revocation cleanup to prevent resource leaks

Cline 4.0.0

Switches to an SDK-backed VS Code extension runtime

Olmoearth Pretrain Adds 1.1

Pretrain dataset release with tokenization config propagation and 2D position encoding refactor

Mistral Common 1.11.5

Mistral Common fixes multi-image content ordering in tokenization and inference workflows

LocalAI 4.5.2

LocalAI 4.5.2 fixes the Opus backend to build and package correctly on macOS/Darwin

Mistral Vibe 2.18.0

Adds clickable URLs in web fetch and search output, context-window token usage display, clipboard image paste on macOS, max-tokens config

Cline CLI 3.0.30

Cline CLI 3.0.30 adds token counting, SAP AI Core provider support, improves model catalog accuracy, and fixes runtime packaging issues

AnythingLLM 1.15.0

AnythingLLM 1.15.0 adds OS-wide Magic Features, dictation, text actions, and autocomplete that work locally across any app without leaving the browser

Claude Agent Sdk Typescript 0.3.193

Browser SDK now supports promptSuggestions option for follow-up suggestions, and Windows subprocess spawning no longer flashes console windows

HelixML 2.11.31

Admins can now grant paid plans directly via wallet PlanOverride without requiring Stripe integration

Olmoearth Pretrain Minimal Publish 0.0.6

Fix literal string to filepath; ---------

Dify 1.15.0

Redesigned onboarding, faster navigation palette, safer deletes with confirmation dialogs

llama.cpp b9788

Fixes quantizing MoE models with mtp

Mistral Common 1.11.4

Adds chat-template integration, deprecates RawAudio in favor of str or bytes, generalizes normalizer aggregation

HelixML 2.11.28

HelixML adds Docker Compose hosting, self-healing health monitoring with restart policies, and DNS proxy fixes for sandbox stability

Ollama 0.30.11

Ollama 0.30.11 fixes GPU detection on hybrid Windows systems, improves speculative decoding, and corrects memory reporting for partially offloaded models

openclaw 2026.6.11-beta.1

Slack/Mattermost relay modes, per-DM model overrides, file-driven agent workflows, safer plugin distribution, and improved reliability across channel delivery and provider integrations

Claude Agent Sdk Typescript 0.3.191

SDK now supports inline diffs in notebook edits, per-model weekly usage limits, and fixes fast mode persistence across conversation turns

LlamaIndex 0.14.23

Dependency bumps across multiple integration directories

Elevenlabs Android 0.10.0

Exposes reconciled message transcript as StateFlow, deprecates text and agent callbacks in favor of event-id variants

Qwen Code 0.19.2

Qwen Code adds image transcription for text-only models, deterministic guards for destructive auto commands, voice dictation over daemon, and MCP resource matching improvements

Jan 0.8.3

Jan 0.8.3 fixes MCP browser port leaks, video upload rendering, SVG markdown handling, and sampler defaults

llama.cpp b9776

Adds LFM2.5-ColBERT-350M and LFM2.5-Embedding-350M support; No impact unless loading those weights

openclaw 2026.6.10

OpenClaw adds automatic fast mode for short conversations, improves model routing reliability across Zai and GLM providers, and fixes session state leaks and policy loss in hook composition

LocalAI 4.5.0

LocalAI 4.5.0 adds depth perception, sound-event tagging, on-device multilingual TTS, PII filtering via NER, speaker-aware realtime conversations, and prefix caching for concurrent multi-user serving without configuration

ComfyUI 0.26.0

ComfyUI adds Qwen3-VL text generation, SCAIL-2 multireference, Boogu image editing, Kling V3-Turbo, Luma Rays 3.2, and job cancellation API endpoints

Client Ts typescript 2.3.0

OpenAPI Doc; Speakeasy CLI 1.763.6 (2.884.13); Typescript v2.3.0

Rslearn 0.1.10

Rslearn adds Sentinel2 cloud filtering, OlmoEarth normalization, and memory-efficient raster operations for remote-sensing machine learning

HelixML 2.11.27

Fixes org visibility, WebSocket sync bugs, prompt-queue ordering, environment variable forwarding under sudo, and disk-pressure detection on non-ZFS hosts

Elevenlabs Python 2.54.0

Python SDK now allows passing custom headers and follow_redirects configuration through client initialization for HTTP request control

Elevenlabs Js 2.54.0

Remove temporary DynamicVariablesConfig fernignore entries

llama.cpp b9767

Improves WebGPU mat-vec path for small-batch decoding

Kimi Code 0.19.1

Fixes ACP editor thread initialization, clears orphaned session data on archive, syncs unread indicators across browser tabs, and refactors web client state management

Qwen Code 0.19.0

Qwen Code 0.19.0 adds voice dictation with native streaming, revivable background agents, artifact publishing for interactive HTML, and fixes loop detection and tool-call guards

langchain== 1.3.11

LangChain 1.3.11 fixes strict tool mode enforcement for non-OpenAI compatible models and updates dependencies

langchain-openai== 1.3.3

Fixes tool schema strictness for OpenAI-compatible models and removes unnecessary response storage fields and stop parameters from API calls

llama.cpp b9760

Relevant if running the server component and permitting file edits, appends at end of file no longer trigger heap overflow

HelixML 2.11.22

AWS Neuron inference support, agent auto-restart, session clearing for backends, and video rendering fixes for Safari 4K and frame ordering

Kimi Cli 1.48.0

Fixes empty reasoning content handling and adds detection to stop tool loops on dead-end streaks

AI release Jun 21, 2026

llama.cpp b9752

Server batch construction refactored for more careful slot handling and improved abort logic

AI release Jun 21, 2026

openclaw 2026.6.9

Telegram messaging gains rich HTML and markdown rendering, agent recovery improves through retry logic and session repair, and official provider plugins become standalone npm packages

AI release Jun 20, 2026

llama.cpp b9739

WebGPU adapter toggles for F16 on Vulkan and NVIDIA. Relevant only if running those backends

AI release Jun 20, 2026

Cline CLI 3.0.29

Cline CLI 3.0.29 hides costs for free models, fixes Z.ai metadata resolution, and reverts model display changes from the previous release

llama.cpp b9732

Server context-shift fix for non-bound n_discard value. Upgrade if running server mode with ctx shifting

Hermes Agent 0.17.0

Hermes Agent 0.17.0 adds iMessage and Raft agent network support, substantially upgrades the desktop app, enables background subagents, and adds image editing capabilities

whisper.cpp 1.9.1

Release: v1.9.1

openclaw 2026.6.8

Telegram and WhatsApp channel delivery now supports rich formatting, agent runs handle state recovery correctly, and model routing adds GLM-5.2 and Claude Haiku 4.5 with safer authentication

langchain-core== 1.4.8

Langchain-core 1.4.8 memoizes tool schemas, caches JSON schemas, preserves token usage in streaming, and drops Python 3.9 support

ComfyUI 0.25.1

ComfyUI added Kling V3-Turbo model support through partner nodes for faster video generation workflows

HelixML 2.11.21

HelixML 2.11.21 fixes streaming frame flicker, improves runner visibility, and refactors agent framework without forking

llama.cpp b9694

Fixed Windows x64 OpenVINO release build link to resolve CI pipeline failures

llama.cpp b9690

Optional USM system allocations for buffers over 1GB if the device supports it

Ollama 0.30.10

Command and Northfamily models now run on Apple Silicon via MLX engine, llama.cpp updated to build 9672

NanoClaw 2.1.17

NanoClaw 2.1.17 adds Codex provider support with per-group selection, fixes provider-error reporting and command-gate access control, auto-upgrades OneCLI gateway on pin changes, and interrupts in-flight turns with slash commands

whisper.cpp 1.9.0

Adds NVIDIA Parakeet model support. Check changelog if running Ruby bindings

HelixML 2.11.20

HelixML 2.11.20 adds disk-pressure admission control for dev containers, fixes orphaned storage cleanup and chat model routing, and corrects Floor node counting logic

Cline CLI 3.0.26

Restored previous model-selection UI expandable picker sections and ClinePass model integration

Maple 3.1.0

Windows desktop port, hardened release provenance verification, paste image support, and improved chat UI for local AI workflows

ComfyUI 0.25.0

ComfyUI adds audio consolidation, 3D preview nodes, partner integrations (Bria, Krea, Gemini), and performance optimizations for inference workflows

openclaw 2026.6.8

GLM-5.2 and Claude Haiku 4.5 support, improved Telegram/WhatsApp rendering, safer model routing, and more reliable agent recovery paths

llama.cpp b9670

Adds Vulkan support for gated delta net with S v=16

whisper.cpp 1.8.7

Drops hardcoded /usr/lib assumption; Relevant if installing to non-standard prefixes

Z Ai Sdk Java 0.3.5

Z AI SDK Java adds async image generation, OCR API, Java 9+ module support, and GLM 5.2 model support

Predicting model behavior before release by simulating deployment

OpenAI released Deployment Simulation to predict model behavior before production using real conversation data, improving safety testing and evaluation accuracy

Ollama 0.30.9

Ollama 0.30.9 adds Cohere2Moe support, fixes token output limits in coding agents, and validates message sizes against context windows

Hugging Face Transformers Patch 5.12.1

Raises PEFT lower bound; Fixes mistral tokenizer resolution when mistral-common is installed

Hugging Face Transformers Patch 5.10.3

Fixes for vLLM integration, token ID handling, InternVL models, and processor offsets in Transformers 5.10.3

Maple Proxy 0.1.10

Maple Proxy v0.1.10; All binaries include

Continue 1.2.24-vscode

Final stable release drops Hub slugs from config templates; Migrate to explicit model definitions if using old syntax

llama.cpp b9642

CUDA operators note F32/F16 restriction on GGML OP REPEAT; KleidiAI macOS build remains disabled

vLLM 0.23.0

DeepSeek-V4 hardening, Model Runner V2 expansion to Llama/Mistral, Rust frontend maturation with streaming and LoRA endpoints, Gemma 4 support, Transformers v5 compatibility, multi-tier KV cache offloading

AI release Jun 14, 2026

llama.cpp b9628

Adds Cohere2MoE vocab for TINY AYA. No impact unless loading those weights

AI release Jun 13, 2026

langchain-openai== 1.3.1

Fixes streamed tool call normalization and structured output fallbacks, adds package version tracking to tracing metadata

vLLM 0.23.0

Minimax M3 models remain unsupported in this version

Ollama 0.30.8

Ollama 0.30.8 improves prompt caching efficiency, fixes provider selection bugs, and hardens MLX inference stability with snapshot creation

Hugging Face Transformers 5.12.0

Adds MiniMax-M3-VL vision-language model support

llama.cpp b9611

Llama.cpp refactored build dependencies to reduce coupling between fit and llama-ext headers

koboldcpp 1.115.1

Koboldcpp 1.115.1 adds Anthropic API support with multimodal and tool calling, Gemma4 UV models, MTP speculative decoding, video generation improvements, and sampling speedups

koboldcpp 1.115

Implements full Anthropic /v1/messages API support including multi-turn conversations

openclaw 2026.6.6

OpenClaw tightens security across transcripts, sandboxing, and tool execution while improving Telegram delivery coherence, iMessage recovery, browser connectivity, and startup latency

langchain== 1.3.8

Langchain 1.3.8 tightens structured output model fallbacks and fixes async middleware decorator typing

LocalAI 4.4.1

LocalAI 4.4.1 restores vLLM 0.22+ compatibility and adds streaming pipeline stages for realtime LLM, TTS, and transcription workflows

Cline 3.89.2

Cline upgrades Anthropic SDK for Node 24 compatibility, fixing provider issues on VS Code 1.123 and later

Cline 3.89.1

Cline restored Anthropic provider support on VS Code 1.123+ and added DeepSeek V4 reasoning format handling

llama.cpp b9594

Refactors normalizer flags into options struct and adds strip-accents support. No model or inference changes

langchain-model-profiles== 0.0.6

Tool call chunk validation added to standard streaming tests, improving LLM integration reliability

langchain-mistralai== 1.1.5

Adds content block token support in callbacks and stabilizes integration tests with rate limiting

LocalAI 4.4.0

LocalAI 4.4.0 adds multimodal capabilities: parakeet.cpp and CrispASR audio backends, video understanding, object detection, LTX-2 video generation, distributed prefix-cache routing, and security hardening

HelixML 2.11.16

HelixML improved agent crash recovery, fixed streaming interrupts, added multi-tenant org redesign with GitHub App bot identity and reporting-line topology, and scaffolded YellowDog compute integration

llama.cpp b9589

CUDA: Fix ssm_scan_f32 data-races missing syncthreads calls to prevent shared memory corruption

langchain-groq== 1.1.3

Langchain-groq 1.1.3 adds content-block-centric streaming v2, strict mode for Groq, standard model properties, and updates dependencies including langsmith 0.8.0 and langchain-core 1.3.3

llm CLI 0.32a3

Tool calls now support pausable chains with unique IDs and human-in-the-loop resumption, enabling approval workflows before model invocation

langchain-openai== 1.3.0

Langchain-openai 1.3.0 adds support for OpenAI's apply_patch built-in tool and updates model profile data

Cline 3.89.0

Cline adds Claude Fable 5 model support, fixes MiniMax M3 thinking controls, and cleans up the Codex model list

llama.cpp b9585

Operators running llama.cpp with plamo2 models should upgrade. Fixes an attention key-value length regression introduced in a prior build

openclaw 2026.6.5

QQBot strips thinking scaffolding, MCP tool results coerce to safe formats, extended-thinking sessions recover after cache expiry, Parallel web-search bundled, Matrix voice-notes and threading improved

AnythingLLM 1.14.0

AnythingLLM now enables tool calling across all providers, adds STT/TTS support, converts web scrapes to markdown, and improves agent summarization

Cline CLI 3.0.21

Cline CLI adds global auto-update control, Vertex AI ADC support with tool use, and fixes session routing bugs plus empty message replay for Bedrock

AI release Jun 8, 2026

llama.cpp b9568

Fixes im2col 1D case for audio models. Relevant only if running audio inference

AI release Jun 8, 2026

langchain-core== 1.4.2

LangChain Core deprecated the problematic dict() method to encourage use of safer serialization patterns in LLM application chains

AI release Jun 7, 2026

Ollama 0.30.7

Ollama adds Hermes Desktop, a native UI for managing Hermes agent conversations and integrations via ollama launch

AI release Jun 7, 2026

llama.cpp b9553

Sampler name matching now case-insensitive and accepts alternative names like top-k alongside canonical top_k format

AI release Jun 6, 2026

llama.cpp b9538

Qwen3.5 video support added; No impact unless loading those weights

Ollama 0.30.6

Ollama adds Gemma 4 QAT models for lower memory use, integrates Oh My Pi coding agent, improves Apple Silicon quantization

llama.cpp b9533

Build system fix resolves compilation failures in llama.cpp, enabling reliable local LLM inference on consumer hardware

langchain-core== 1.4.1

Removes Bedrock prevalidation from load path; Bumps uuid-utils to 0.16.0

vLLM 0.22.1

VLLM 0.22.1 adds Mellum v2 support, enables zentorch-accelerated quantized inference on AMD Zen CPUs, and fixes multi-node Ray serving hangs and model initialization regressions

Cline CLI 3.0.20

Plugin wrappers now use readable source names instead of hashes, making installed extensions easier to identify and manage

AI release Jun 4, 2026

Hugging Face Transformers Patch 5.10.2

Fixes model conversion bug for CLIP-related models including SAM3. Update if those weights are in use

AI release Jun 4, 2026

llama.cpp b9498

WebGPU FlashAttention refactored with standardized quantization support, relevant if running WebGPU backend

AI release Jun 4, 2026

Maple Proxy 0.1.9

Maple Proxy v0.1.9; All binaries include

openclaw 2026.6.1

Agents now recover cleanly from interrupted tool calls and stale bindings, channels stabilize across messaging platforms, and Skill Workshop adds governed proposal review with CLI and UI tooling

ComfyUI 0.24.0

Fixes dtype handling for birefnet and dino3 models, upgrade if using those

Cline 3.87.0

Cline adds MiniMax M3 model support and patches security vulnerabilities in VS Code extension dependencies including xmldom, axios, and undici

langchain-deepseek== 1.1.0

Content-block streaming v2 support, dependency updates including langchain-core 1.3.3, and Azure endpoint handling fixes

Ollama 0.30.3

Ollama added Gemma 4 12B model support, expanding locally-runnable open-weight options for builders

Hugging Face Transformers 5.10.1

5.10.0 was yanked; Upgrade directly to 5.10.1

HelixML 2.11.15

HelixML fixes nested GPU passthrough for containerized desktops, requires explicit org_id for credit grants, and improves chat message pacing to model realistic user behavior

llama.cpp b9486

Adds support for qwen3 SSM architectures

Ollama 0.30.2

Auto-install Cline CLI; Qwen code integration added; Template logging improved for troubleshooting

langchain== 1.3.4

LangChain 1.3.4 improves human-in-the-loop rejection guidance for better control flow in AI agent workflows

llamafile 0.10.3

Fixes SIGSEGV regression in GPU initialization, restoring CPU fallback when GPU setup fails

langchain== 1.3.3

LangChain 1.3.3 adds subagent run projection to typed channels and human-in-the-loop interrupt predicates for agentic workflows

llama.cpp b9473

SWA checkpoint optimization reduces KV cache storage masked cells, improving memory efficiency for inference

whisper.cpp 1.8.6

Whisper.cpp 1.8.6 refactors ffmpeg transcoding logic and fixes CI header path resolution

Open WebUI 0.9.6

Open WebUI 0.9.6 adds incremental knowledge base sync from multiple sources, filesystem commands for AI models to search KB contents, and nested folder organization for managing large document collections

ComfyUI 0.23.0

MediaPipe face detection, multi-threaded model loading for faster startup, OAuth 2.1 support, and logic nodes shipped to improve performance and API standardization

Jan 0.8.2

Jan 0.8.2 adds pause/resume for model downloads, AMD ROCm support, persistent assistant parameters, and fixes context limits, model crashes, and UI responsiveness