vllm-project / vllm-project/agentic-api
Add deployment, authentication, and persistence end-to-end verification
Nobody has claimed this yet.
- Dominant language
- Rust
- Stars
- 284
- Forks
- 74
- Avg merge
- 1d 17h
- Merged PRs (30d)
- 93
Description
Parent: #102
Create a verification matrix for the hosted deployment architecture.
Scope:
- Container startup and health/readiness checks.
- PostgreSQL initialization and restart persistence.
- OIDC/GitHub login and invalid-token rejection.
- API-key creation, use, scope enforcement, expiration, and revocation.
- Cross-user/tenant isolation.
- HTTP streaming and WebSocket behavior behind a reverse proxy/ingress.
- Secret leakage and log-safety checks.
Acceptance: the deployment can be validated repeatably before release on both a local container stack and at least one hosted target.
See the parent issue: https://github.com/vllm-project/agentic-api/issues/102
Enterprise Readiness
Extend this deployment verification work into a repeatable release qualification matrix for standalone and generic reverse-proxy/ingress deployments of Agentic API.
Additional acceptance criteria:
- Run the actual Agentic API server with deterministic inference/tool servers and a real supported database; exercise Responses JSON/SSE/WebSocket plus native Messages JSON/SSE and count_tokens where supported.
- Cover continuation, explicit and automatic compaction, client-executed function tools, built-in tools, approval resume when available, and mixed tool calls.
- Verify one execution per scheduled call ID within an active turn, deterministic stored-item appends, one public terminal lifecycle, and no automatic replay after an ambiguous tool outcome.
- Exercise credential separation, tenant isolation, resource-owner routing, WebSocket affinity, restarts, slow consumers, disconnects, upstream errors, and partial usage accounting.
- Publish a capability/result matrix that distinguishes deterministic mock tests, cassette replay, and live engine coverage; unsupported cases are explicit rather than counted as passes.
- Extend the existing recorder workflows and #211 engine-conformance program, including existing Dynamo Messages recordings from #292. Pin engine version, model, parser configuration, protocol, and transport.
This issue owns deployment-level qualification; #211 owns the engine recording/live-matrix program. Existing parent #102 remains relevant.
Enterprise Readiness tracker: https://github.com/vllm-project/agentic-api/issues/316
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start with parent #102 and the existing recorder workflows, then review the boundary with #211, the Dynamo Messages recordings from #292, and the Enterprise Readiness tracker #316. Define repeatable local-container and hosted-target runs covering the listed protocols and deployment behaviors. Done means a published capability/result matrix with pinned configuration and unsupported cases explicitly identified rather than counted as passes.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- github, postgresql, rust
- Domain
- authentication, backend-api-design, cloud, databases, devops, release, security, testing
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Quiet
- Clarity
- Mostly clear
- Newbie friendliness
- 25/100