Agents & MCPGitHub · Pydantic AI
Pydantic AI: CLAI2 bleeding
Sdists rebuilt from main for /update on the CLAI bleeding channel. Not a release.
kind:agentrepo:pydantic/pydantic-aiproject:Pydantic AI
Desk
Top repos GitHub is watching today, curated agent/MCP/RAG releases, and official blog.
Top today
15
Trending
1005
Releases
2962
Blog
69
Sdists rebuilt from main for /update on the CLAI bleeding channel. Not a release.
FRV repo UI shards run WebKit-owned suites, so provision WebKit alongside the cached Chromium install. Co-authored-by: roboclaw-bot [email protected] Co-authored-by: RomneyDa [email protected]
What's changed Added permissions.blockReadsExemptPaths to the Settings type: paths, read only from --settings and settings the host sends, that blockReadsOutsideWorkingDirectories lets through without making them working directories Updated to parity with Claude Code v2.1.316 Update npm install @anthropic-ai/[email protected] # or yarn add @anthropic-ai/[email protected] # or pnpm add @anthropic-ai/[email protected] # or bun add @anthropic-ai/[email protected]
Minimal CEF distribution for the macOS desktop app, built with proprietary_codecs=true and ffmpeg_branding=Chrome. sha256: d5605e50a736b8e107a58ea900d63e6893d4a66d00dda580976378f4de5d52d2
What's Changed Route a channel with many long member descriptions by @Sniezka1927 in #4102 Say that the OpenBot source code is open, not that OpenBot is open source by @NorbertBodziony in #4104 Explain why a Workspace only Claude agent cannot start its sandbox by @Sniezka1927 in #4108 Build the macOS ARM64 release on macos-15 by @Sniezka1927 in #4109 Drop a dragged sidebar chat in the section under the pointer by @Sniezka1927 in #4112 Use the installed provider CLI and keep one downloaded versio
Changes since 1.2.18 release(langgraph): 1.2.19 ( #9504 ) release(checkpoint-sqlite): 3.1.3 ( #9505 ) fix(langgraph): keep trace_policy from rewriting streamed node payloads ( #9486 ) fix(langgraph): give messages and tools events the namespace of their graph ( #9458 ) chore(deps): bump setuptools from 83.0.0 to 84.1.0 in /libs/langgraph ( #9433 ) fix(langgraph): emit node-output messages from a graph streamed inside a node ( #9463 ) fix(langgraph): respect subgraphs=False in RemoteGraph.stream
1.24.0 - 2026-10-30 Warning Redis context index migration: The beta RedisContextProvider now preserves exact scope identity and rejects the legacy raw-scope index schema. Delete the old context index and its context hashes, let the provider create a new index, and re-ingest the context data. overwrite_index=True does not migrate incompatible indexes. Important Python support: Python 3.11 or later is now the supported baseline, subject to package-specific upper limits. Python 3.10 installation re
harness-typescript/v0.3.3 Auto-drafted from commits in harness-typescript/v0.3.2..harness-typescript/v0.3.3 , grouped by conventional-commit type. Edit on the release page after publish if you want a polished writeup; the canonical release notes live on the website. 🚀 Features feat(models): default Bedrock maxTokens to the model info output limit (Python + TypeScript) ( #5645 ) ( 0a50a80 ) feat(models): add Model.model_info with capabilities and limits (Python) ( #5627 ) ( bcd704b ) 🐛 Fixes fix(
My Work: refresh stays on your agent, and a Feedback button ( #5190 ) When My Work's automatic refresh got an answer from Claude Code that claude-mem refused (most often a review over 1,200 characters), the whole refresh moved on to the next logged-in agent. On machines with Codex installed, that meant hidden codex exec --ephemeral runs on the owner's ChatGPT account, even with CLAUDE_MEM_PROVIDER=claude ( #5189 : about 230 runs in four days). A refused answer now ends that refresh on the agent
Adds prepared listing JSON to avoid repeated decoding and rebuilding of server payloads on listing reads. Prepared reads are disabled by default and support bounded backfill and flag rollback. Listing rows release their database connection before JSON processing; pool defaults remain 60/10. Focused response/update/delete/backfill/rollback checks passed. Local encoding benchmarks over saved public records measured 63% less encoding time and 88% fewer allocations. These are encoding-cost results,
Release coding-agents v0.8.4
1.123.91 (2026-10-30) Bug Fixes core: Fix $('Node') expressions returning null for legacy Code node output ( #42249 ) ( 18c89cd )
Hermes Agent canary 20261030T064851Z
Qwen CUA Driver prebuilt binaries (vendored under packages/cua-driver ). macOS : codesigned + notarized universal binary + QwenCuaDriver.app Linux : unsigned (x86_64 + arm64, glibc 2.31 floor) Windows : unsigned UIAccess worker + native SDK payload (x86_64 + arm64); deployments must sign and trust the worker on the target machine before UIAccess activation Node.js : the same workflow publishes @qwen-code/cua-sdk after a clean install against these release assets and publishes the same-version br
Added The BYOK selector can ask a custom endpoint what it serves ( #1543 , externally contributed by vito-bush ). POST /api/v1/llm/credentials/models lists the chat-capable models an OpenAI-compatible endpoint publishes at GET {base}/models , so the custom provider's model no longer has to be a hand-typed id — closing the gap the #1481 review named. Bounded (10 s whole-exchange deadline, 2 MB decoded cap, 500 models returned), redirects refused, the air-gap and SSRF guards applied before anythin
Release version 5.7.21
Internal Testing Build This is an automated canary build from main branch. Commit: 150e8da Short SHA: 150e8da Built: 2026-10-30T00:02:25Z ⚠️ This build is for internal testing only and may be unstable.
Minor Changes #2771 54ae3e8 Thanks @mattzcarey ! - Breaking: Think reloads getSkills() sources every turn, so skills.workspace(this.workspace) picks up skills written to /workspace/.agents/skills ; an overridden getSystemPrompt() now always renders above context blocks instead of being dropped once any block exists; skill catalog changes reach the model as a context reminder rather than rebuilding the cached system prompt; skillWorkspace and the extension "skill" context block type are removed.
What's Changed Features feat(ingestion): attribute AI gateway spans to their gateway API key and service by @wochinge in #19398 feat(telemetry): report AI gateway request counts from self-hosted instances by @wochinge in #19433 feat(pagerduty): send alerts, project notifications and prompt changes to PagerDuty, with the routing key on the automation by @marksalpeter in #19411 feat(evals): add filter conditions for trace-level by @marliessophie in #19400 feat(ai-gateway): add an analytics tab to