Run The Files API¶
The independent demo/files_api project exposes an OpenAI-compatible Files API
over S3-compatible storage. It has its own package, lockfile, Dockerfile, image,
settings, and tests. It does not depend on langgraph-openai-serve, LangGraph,
or the demo graph API.
Run The Service¶
Configure the DEMO_API_FILES_* values in demo/.env, then choose one mode:
The OpenAI base URL is http://localhost:3006/v1; health is available at
http://localhost:3006/health.
The service supports upload, list, retrieve, content, and delete operations for
purpose="user_data". The maintained UIs reach it through Bifrost's
lgos-files provider or LiteLLM's litellm_proxy Files provider; they do not
connect directly. Direct protocol checks may call it on port 3006. The
file-input graph uses the same service to resolve an incoming file_id.
Use A Custom Repository¶
Implement the structural FileRepository protocol in
lgos_files_api.contracts and pass it to the public create_files_app
factory:
from lgos_files_api import create_files_app
from my_repository import MyFileRepository
app = create_files_app(MyFileRepository())
The default command uses create_s3_app; a custom deployment supplies its own
ASGI module and storage lifecycle. Neither inheritance nor a dependency on
LGOS or LangGraph is required.
Ownership Boundary¶
Only this service receives the DEMO_API_FILES_AWS_* credentials and bucket
configuration. LGOS API replicas receive only the Files base URL needed by a
graph that resolves IDs. Chainlit's native element-storage credentials remain
separate, and Open WebUI retains its native raw upload copy in its own data
volume.
The demo is a deliberately small reference implementation. Before exposing it to untrusted clients, add authentication, tenant isolation, upload limits, retention, malware policy, and the availability controls required by the deployment. See Accept And Display Files for the client and graph flow and Settings and Commands for configuration.