Scenarios

Copy-paste, end-to-end recipes. Each one runs against a real proxy — no hand-waving.

A "scenario" here is a concrete, runnable request/response shape — not just prose. Most of these are backed one-to-one by an integration test in internal/server/scenarios_test.go (token auth, auto routing, concrete bypass, streaming heartbeat, pull-worker); dynamic-group routing and conversation-mapped agent sessions are covered by dedicated tests elsewhere in the suite (server_test.go, backend/devin_test.go). Same requests, same expected headers/status codes as the tests — this page stays honest as the code changes.

Pick a scenario

• 1 · Route automatically — model:"auto" classifies the task and picks a tier.
• 2 · Bypass to a concrete model — name a real model, routing is skipped.
• 3 · Token-based auth, end to end — generate a token, wire it in, 401 → 200.
• 4 · mTLS — full CA + server + client chain, client certs required.
• 5 · Agent session that remembers — same conversation id resumes a real Devin session.
• 6 · Pull-worker — a remote agent registers a queue and answers requests.
• 7 · Streaming with heartbeat — SSE chunks plus : ping keep-alives.
• 8 · Dynamic group routing — a named group with levels, classified per request.
• 9 · Secrets via .env layering — precedence across process env and config files.

← Back to docs overview