This release includes 2 security fixes for security teams reviewing exposed deployments.
Topics
Affected surfaces
Summary
AI summaryUpdates Highlights, HR/legal/finance, and Syft across a mixed release.
Full changelog
Highlights
LLM_PROVIDER=mockfully works end-to-end.BaseAgentno longer crashes callingbind_tools()on the mock chat model, and the mock's canned LLM responses now match the real pipeline's call sequence — a fresh clone +uv sync+LLM_PROVIDER=mockgives a coherentPOST /runresponse with zero cost and no API key.cp .env.example .envalso no longer crashes on the emptyPACK_DEFAULT_BUDGET_USDdefault.- Current-generation LLM model defaults across every provider: Anthropic
claude-sonnet-5, OpenAIgpt-5.5, Googlegemini-3.5-flash, Bedrockanthropic.claude-sonnet-5(config,.env.example, and thecore/cost.pypricing table). api/main.pysplit into modules —api/app.py,lifespan.py,middleware.py,dependencies.py,router_factory.py,api/endpoints/*— instead of one monolithic file.- 13 domain packs, namespaced by business domain (
research/,productivity/,hr/,finance/,legal/), each with typed request/response schemas and auto-generated API routes. Regulated verticals (HR/legal/finance) ship compliance guardrails and are gated behindREGULATED_PACKS_ENABLED=false(HTTP 403 until you opt in after completing the pack'sCOMPLIANCE.md). - Security hardening: Security Scanning workflow green on
main(gitleaks allowlist for confirmed false positives, patchedstarlette/langsmithCVEs), plus the existing rate limiting, input validation, and prompt-injection guards. - Supply chain: GHCR images signed with Cosign and published with an SBOM (Syft) on every
mainbuild. - README rewritten for public launch: an honest "Why this instead of…" section (vs. LangGraph Platform, vs. generic agent-service-toolkit templates), a bulletproof quickstart (mock-provider callout,
uvinstall link, cost estimate, mock-search note), and credibility signals (758+ tests, ~87% coverage, scaling guidance, PR template,SECURITY.md).
Upgrading
No breaking config changes since 0.5.0 for existing deployments. If you pin ANTHROPIC_MODEL / OPENAI_MODEL / GOOGLE_MODEL / BEDROCK_MODEL explicitly in your .env, your override is unaffected — only the defaults changed.
See CHANGELOG.md for the full list of changes.
Security Fixes
- Patched `starlette` and `langsmith` CVEs; Security Scanning workflow now green on `main`
- Added gitleaks allowlist for confirmed false positives
Weekly OSS security release digest.
The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.
No spam, unsubscribe anytime.
Share this release
About Production-grade LangGraph template
All releases →Related context
Related tools
Beta — feedback welcome: [email protected]