LLM Proxy for Teams

The multi-developer problem

When one person uses an LLM API, cost tracking is simple. When five developers share an API key, you have no idea who spent what. When engineering, product, and QA all call the same models, you cannot attribute costs to projects or enforce limits per team.

An LLM proxy gives every team their own API key, their own spend limits, and their own request logs, all through a single shared endpoint.

How it works

# Create a team
curl -X POST http://localhost:4200/api/teams \
  -d '{"name": "backend", "daily_limit_usd": 50}'

# Generate a team API key
curl -X POST http://localhost:4200/api/teams/backend/keys
# sk-sy-team-backend-a8f21c4e...

# Each team uses their own key
export OPENAI_API_KEY=sk-sy-team-backend-a8f21c4e...
export OPENAI_BASE_URL=http://your-proxy:4200/v1

Per-team isolation

Separate spend tracking. Each team's costs are tracked independently. The cost dashboard shows per-team breakdowns by model, provider, and day.

Spend limits. Set daily and monthly caps per team. When a team hits their limit, requests are blocked with a clear error, not silently dropped.

Rate limiting. Prevent one team from monopolizing the API. Set per-team RPM (requests per minute) and TPM (tokens per minute) limits.

Isolated logs. Each team sees only their own requests in the dashboard. No cross-team data leakage.

Audit trail

Every request through the proxy is logged with the team, the user (if keys are per-person), the model, the cost, and the full request/response. Stockyard's Brand audit ledger creates a SHA-256 hash-chained record that is tamper-evident. For compliance reviews, you can export evidence packs showing exactly who used what, when, and how much it cost.

No shared infrastructure to manage

The proxy runs as a single binary with embedded SQLite. No Postgres cluster for the team to share, no Redis for rate limits. One file, one process. Back it up by copying the database file.