Internal analytics dashboard for New API usage logs. The app reads PostgreSQL log data and visualizes calls, token consumption, model usage, channel usage, per-user aggregation, and raw request logs.

Requirements

Bun for local development and tests.
Node.js 20 for the production Docker image.
PostgreSQL database with New API tables described in docs/database.md.
PG_CONNECTION_STRING set to a PostgreSQL connection string.

Local Development

bun install
bun run dev

Open http://localhost:3000.

Verification

bun test
bun run lint
bun run build

bun test covers parser helpers, metric conversion, query behavior, cache behavior, theme helpers, and selected API route guardrails.

Smoke test: AO/Codex can update this README, pass lint, and open a GitLab MR.

Configuration

Create an environment file or export:

PG_CONNECTION_STRING=postgres://user:password@host:5432/database

The app uses this variable in lib/db.ts to create a pg connection pool.

OIDC authentication is optional. If no OIDC variables are set, the dashboard and API remain open. To require login through a standard OIDC provider such as Sinodoor, add:

OIDC_ISSUER=https://casdoor.example.com
OIDC_CLIENT_ID=analytics
OIDC_CLIENT_SECRET=replace-me
AUTH_SECRET=replace-with-random-secret
# Optional login button label:
OIDC_PROVIDER_NAME=Sinodoor

Generate AUTH_SECRET with a stable random value, for example:

openssl rand -base64 32

When OIDC is enabled, configure the provider redirect URI as:

https://your-analytics-domain/api/auth/callback/oidc

Partial OIDC configuration is treated as an error instead of falling back to open access.

Deployment

The included Dockerfile builds a standalone Next.js output and starts server.js on port 8019.

docker compose up -d --build

docker-compose.yml expects .env.production and an external Docker network named sinobridge.

Operational Notes

API query parameters are validated at the route layer. Invalid integers or reversed date ranges return HTTP 400.
Ranking limits are capped at 100 rows per request.
Log page size is capped at 200 rows per request.
Query results are cached in-process for 120 seconds with a 500-entry cap.
Metric definitions and quota conversion are documented in docs/metrics.md.