Demo Settings And Commands¶

This reference describes the independently locked projects and Compose stack under demo/. These commands and DEMO_* settings are not part of the langgraph-openai-serve package API.

demo/.env.example is the source of truth for demo environment values. Copy it to demo/.env and customize it before starting services or running live integration tests. Just loads it when present; exported environment variables take precedence. Local tests, lint, formatting, type checks, and dependency synchronization can run without it. This reference explains settings without duplicating their defaults.

Projects¶

Each Python project has its own pyproject.toml, virtual environment, and uv.lock; demo/ deliberately is not a uv workspace.

Common Commands¶

Run these from the repository root. Configure demo/.env for service and integration commands:

Common service names are lgos-db, lgos-demo-api, lgos-api-coding-agent, lgos-background-worker, lgos-files-api, lgos-postgres-mcp, lgos-bifrost, lgos-litellm, lgos-chainlit, and lgos-openwebui. Put arguments for the underlying command after -- when a recipe has its own options, for example just demo/test --editable -- -q.

Use just demo/ to list recipes in the local, integration, checks, docker prod, and docker dev groups. Docker recipes use published images by default, except the coding-agent service, which always builds locally. --dev selects checkout builds and editable LGOS for both graph API projects. Shared recipes appear in both Docker groups.

just --usage demo/api shows its options and defaults. The --port option overrides the dotenv value; exported variables work too, for example LGOS_DEMO_API_PORT=3104 just demo/api. Add Just's --dry-run before the recipe to inspect commands. To validate Compose using the template without creating an environment file, check both gateways in published, development, and telemetry configurations:

just --dotenv-path demo/.env.example demo/compose-config

The local Chainlit process uses the host-reachable DEMO_GATEWAY_HOST_URL. Open WebUI always runs its pinned published image. See the Chainlit and Open WebUI guides.

just demo/test-litellm --editable and just demo/test-bifrost --editable run the focused OpenAI SDK checks. just demo/test-background-gateway --editable selects the route and model from OPENAI_GATEWAY_TYPE; use --base-url and --model to override them. OPENAI_GATEWAY_TYPE=litellm|bifrost selects the gateway used by both maintained UIs. Responses and Files use its normal managed/native routes. LiteLLM metadata comes from native /model/info after model sync; Bifrost metadata comes from native /v1/models after catalog sync.

Stack Settings¶

Integration Test Settings¶

Live test recipes read their endpoints from the DEMO_TEST_* values in demo/.env.example. Their native options can override those defaults:

just demo/test-postgres --uri postgresql://lgos:lgos@localhost:5432/lgos --editable
just demo/test-direct --base-url http://localhost:3104/v1 --files-url http://localhost:3106/v1
just demo/test-litellm --base-url https://litellm.example.com/v1 -- --verbose

Use just --usage demo/test-bifrost for the normalized and catalog endpoint options. --editable overlays the parent LGOS checkout; arguments after -- go to pytest. CI can export DEMO_API_TEST_POSTGRES_URI and run just demo/test-postgres --editable without a dotenv file.

OpenTelemetry Settings¶

These settings apply when using just demo/compose --otel, optionally together with --dev:

The OTEL overlay uses the OpenTelemetry always_on sampler, so application traces are exported without SDK sampling. The selected remote backend owns retention.

The OTEL overlay requires both OTEL_COLLECTOR_GATEWAY_ENDPOINT and OTEL_HOST_NAME; set them per machine in demo/.env. The endpoint URL scheme controls transport security: use https:// for TLS and http:// only when the gateway intentionally accepts cleartext OTLP/HTTP.

Demo API Settings¶

The demo API and its worker run lgos serve and lgos worker. Their server settings, including LGOS_POSTGRES_URI, LGOS_INTERRUPT_TTL_MINUTES, LGOS_BACKGROUND, LGOS_HATCHET_WORKER_SLOTS, and LGOS_CORS_ORIGINS, are described in Run The LGOS Server. The graphs share OPENAI_GATEWAY_BASE_URL with the UIs and call it with DEMO_GRAPH_GATEWAY_API_KEY, which defaults to the UIs' key. Model calls use /v1. The local API, worker, and notebook recipes set the gateway root to DEMO_GATEWAY_HOST_URL and the key to DEMO_GRAPH_GATEWAY_API_KEY.

Graph-specific settings use the DEMO_API_ prefix:

The background demo uses the create_hatchet_task defaults: 30-minute schedule and one-hour execution timeouts with no retries. Hatchet's data retention decides how long Responses stay retrievable. Self-hosted deployments should inject any additional native HATCHET_CLIENT_* connection settings into both the API and worker processes.

The API serves the OpenAI routes at the package's default /v1 prefix, which the gateway wiring assumes. It also reads the package-owned LGOS_OPENAI_API_DOCS_ENABLED and LGOS_ENABLE_LANGFUSE settings documented in the package Reference. Its settings model supports a local .env file; the installed LGOS package itself reads only process environment values or explicit constructor arguments.

Coding Agent Settings¶

These settings belong to demo/api-coding-agent. Compose derives the base URL and key from the gateway settings; the other defaults live in demo/.env.example.

See the coding-agent graph guide for its shared workspace, container execution boundary, and invocation examples.

Files API Settings¶

These settings belong only to the independent demo/files_api project.

Open WebUI Settings¶

These settings configure the Open WebUI container. Its Generic Pipe also reads the shared gateway values under Stack Settings from its environment.

See Chainlit settings, Open WebUI setup, and the example graph catalog for component-specific details.