v0.3.1 — Request tracing & accounting fixes

🚀 Get started

New to AISIX? Get the gateway running and route your first LLM request in minutes:

• 📖 Documentation — https://docs.api7.ai/ai-gateway/
• ⚡ Self-hosted quickstart (gateway + etcd via Docker Compose) — https://docs.api7.ai/ai-gateway/getting-started/self-hosted-quickstart

📦 Download

Pull the container image from the GitHub Container Registry:

docker pull ghcr.io/api7/aisix:0.3.1

Also mirrored to docker.io/api7/aisix for private/offline deployments. All tags are listed on the package page.

---

A patch release on top of v0.3.0: request tracing across every response, tighter token accounting so rate limits can't be gamed, and version self-reporting so operators can pin a running node to its exact image and commit.

🔭 Observability

• Request tracing — x-aisix-request-id is stamped on every proxy response across the whole handler family, so any response correlates with its access log and usage event (#737)
• Self-reporting version — the release version is stamped into the binary at build time; the gateway reports <version>+sha-<short> via --version, the Server header, and the managed-mode heartbeat, so operators can match any running node to its exact image and commit (#736)
• New cache-inclusive token_type=total on the tokens_by_client metric (#740); cooldown metrics emitted on deployment-state transitions (#738)

📊 Accounting

• Cache-token accounting — Anthropic cache tokens are counted in TPM commits on /v1/messages and /v1/responses, so rate limits can't be bypassed via cache reads (#733)
• Telemetry robustness — batch cost-attribution events now always carry a UUID request_id, fixing a telemetry-flush poisoning bug (#734)

Full changelog: v0.3.0...v0.3.1