v0.9.0: chore(release): bump version to 0.9.0 (#1855)

What's Changed

Dependencies

• chore(deps): bump praxis-proxy/conventions/.github/actions/coverage-check from 0.3.0 to 0.4.0 by @dependabot[bot] in #1844
• chore(deps): bump astral-sh/setup-uv from 10.3.0 to 10.3.2 by @dependabot[bot] in #1841
• chore(deps): bump the codeql-action group with 2 updates by @dependabot[bot] in #1840
• chore(deps): bump praxis-proxy/conventions/.github/actions/supply-chain-audit from e1d0ff870a1abb514acc3616624824773cc767b6 to 1a54eaafe74b88fbcc86d37670be340605273486 by @dependabot[bot] in #1842
• chore(deps): bump docker/login-action from 4.6.0 to 4.7.0 by @dependabot[bot] in #1843
• chore(deps): bump the rust-dependencies group with 14 updates by @dependabot[bot] in #1845

Other

• fix(anthropic): accept nullable custom tool fields by @mkoushni in #1805
• fix(anthropic): reject lossy streaming outcomes by @mkoushni in #1803
• docs(cli-vllm): reconcile the client guide with local deployment lessons by @nathan-weinberg in #1795
• chore(deps): bump ctutils to 0.4.5 (RUSTSEC-2026-0385) by @leseb in #1814
• fix(responses): compact reasoning and validate web-search events by @leseb in #1813
• fix(responses): validate web-search queries before dispatch by @nathan-weinberg in #1781
• feat(pi): support the pi coding harness alongside Claude Code, Codex, and OpenCode by @nathan-weinberg in #1810
• fix(openai): lower orphan function_call_output items to Chat user text by @skamenan7 in #1733
• fix(compact): accept schema-valid empty conversations in explicit compact by @Artemon-line in #1804
• fix(ci): use CUDA 13 vLLM acceptance image by @leseb in #1819
• fix(deps): update zstd crates past the October advisories by @cdoern in #1820
• feat(store): response-store TTL and background reaper (#630) by @leseb in #1811
• ci(container): publish experimental image by @derekhiggins in #1817
• ci: decouple compile-bound jobs into parallel lanes (~38m → ~26m) by @leseb in #1815
• fix(responses): publish stored reasoning defaults into the live request body by @skamenan7 in #1793
• test(vllm): stabilize OpenCode 2.x GPU acceptance by @leseb in #1832
• test(pins): single-source Codex and Claude Code pins by @nathan-weinberg in #1827
• feat(crypto): add runtime cryptographic readiness checks by @rhdedgar in #1588
• test(routing): run non-UTF-8 projected secret test on Linux only by @esnible in #1836
• docs(intelligent_route): update routing and overlay reference by @nerdalert in #1837
• perf(gemini): move owned request fields and generated JSON values by @skamenan7 in #1830
• build(deps): bump pulldown-cmark to 0.13.5 (RUSTSEC-2026-0414) by @leseb in #1853
• test(anthropic): add SDK v1 and v2 Messages compatibility lanes by @Artemon-line in #1838
• fix(guardrails): Add service_account_token filter (destination-bound … by @christinaexyou in #1507
• feat(xtask): port the pins job to cargo xtask check-pins by @nathan-weinberg in #1828
• feat(store): add atomic response-object compare-and-swap by @cdoern in #1741
• fix(anthropic): improve Chat Completions request compatibility by @nerdalert in #1851
• chore(release): bump version to 0.9.0 by @leseb in #1855

New Contributors

• @esnible made their first contribution in #1836

Full Changelog: v0.8.0...v0.9.0

Release CI evidence

Release commit: e88829542814df6f4496c2cc5e30f0d5072b313e. Snapshot: 2026-10-30T14:52:59.530711+00:00.

Results below record CI jobs and steps. A successful workflow can include skipped tests; job counts are not test counts.

CI suite	Scope	Result	Jobs / missing evidence	Run
Release test gate	Unit, schema, and integration test steps	success	1 success	run 41059494049, attempt 2
Tests	Rust unit tests, lint, feature matrix, and report checks	success	20 success	run 41059282706, attempt 1
Integration	Schema, inference fixtures, integration, and SDK/client tests	success	1 skipped, 19 success	run 41059282473, attempt 1
PostgreSQL store	PostgreSQL unit and integration tests	success	1 skipped, 2 success	run 41059282673, attempt 1
Valkey rate limits	Valkey backend unit and integration tests	success	1 skipped, 2 success	run 41059282888, attempt 1
vLLM integration	Simulator, live CPU, and optional GPU/client jobs	success	9 skipped, 4 success	run 41059282532, attempt 1
Anthropic vision	Anthropic SDK image requests on GPU	not_run	No eligible workflow run for the release commit	Unavailable
OpenAI conformance	Pinned schema and conformance checks	success	1 skipped, 1 success	run 41059282555, attempt 1
OpenResponses conformance	Responses translation against CPU vLLM	success	1 skipped, 1 success	run 41059282654, attempt 1
Container	Container build and runtime smoke tests	success	2 success	run 41059282755, attempt 1
vLLM GPU images	Model image builds and smoke tests	not_run	No eligible workflow run for the release commit	Unavailable
FIPS	UBI image checks and FIPS host proof	success	2 success	run 41059282677, attempt 1
Coverage	Instrumented Rust tests	success	1 success	run 41059282627, attempt 1
MSRV	Minimum supported Rust compilation	success	1 success	run 41059282858, attempt 1
Supply chain	Dependency, license, and vulnerability checks	success	2 success	run 41059282802, attempt 1
CodeQL	Static security analysis	success	2 success	run 41059282509, attempt 1
Documentation	Rustdoc checks	success	1 success	run 41059282698, attempt 1
Unicode safety	Hidden Unicode checks	not_run	No eligible workflow run for the release commit	Unavailable
Nightly	Scheduled lint, documentation, tests, and audit	not_run	No eligible workflow run for the release commit	Unavailable

See attached release-qualification.md for job and step outcomes, release-qualification.json for machine-readable CI and GPU evidence, and qualification.json for the original GPU report.

Selection: Newest eligible run per workflow for the exact release commit; latest attempt only. The current release run is pinned when supplied. No older green fallback.

• CI records workflow head SHA and job/step outcomes, not individual test counts or complete API conformance.
• PR runs are excluded because their checkout can be a synthetic merge; no evidence is borrowed from other commits.
• Missing runs may reflect path filters or optional triggers; no execution is inferred.
• This is a publication-time snapshot; queued or running jobs may finish later. Rerun Release to refresh it.
• GPU qualification identifies its tested checkout and binary separately from CI and the published container.

GPU evidence: unavailable. no completed Responses GPU attempt for this commit.
To collect GPU evidence, dispatch vllm-integration.yaml with run_live_vllm=true on this release tag, wait for its summary, then rerun Release on the same tag.

vLLM Responses gateway qualification

vLLM Responses gateway qualification: unavailable for this release commit. No usable exact-commit GPU qualification report was found.

Reason: no completed Responses GPU attempt for this commit.

Limitations: selected text and tool behavior over native Responses and Responses-to-Chat HTTP/SSE paths only; streamed Conversation append and multimodal/background/WebSocket paths are excluded. See the attached qualification.json for case reasons and separate client/tool results.