serve server.http routes through the agent data plane - #6918
Draft
theomonnom wants to merge 3 commits into
Draft
theomonnom wants to merge 3 commits into
theomonnom wants to merge 3 commits into
Conversation
AgentServer.http is a FastAPI app, so agents can serve their own routes
alongside the job worker:
@server.http.get("/token")
async def token() -> dict:
return {"ok": True}
Users get pydantic validation, dependencies, middleware, websockets and
OpenAPI without the framework reimplementing any of it.
The health check and worker info are now ordinary routes on that app,
registered after the user's decorators. A health route of your own keeps
GET /, and agent_health() exposes the checks it would have run. GET
/worker stays reserved, since the control plane and the lk CLI read it.
uvicorn replaces the aiohttp runner. It captures SIGINT and SIGTERM
unconditionally, so capture_signals is overridden to leave them to the
CLI, and its own loggers are quieted because the startup banner
duplicates ours. The bound port is read back from the IPv4 socket, since
an empty host binds one socket per address family, each with its own
port when the port is 0.
websockets is a new dependency: uvicorn refuses websocket upgrades
without an implementation installed, and it only reached the tree
through optional plugins.
User routes declared on server.http are served at /agents/{deployment}/{path}
through the LiveKit server: the worker registers a route manifest on the
control connection and dials a fixed pool of wires speaking AgentHttp.Frame;
each stream bridges one opaque HTTP/1.1 exchange into the FastAPI app
in-process (h11 -> ASGI, two-level credit flow control, per-wire parameters
from the attach response, redial within the epoch). No local listener serves
user routes; a health-only listener remains for probes.
longcw
force-pushed
the
longc/server-http
branch
from
September 16, 2026 04:46
cf30080 to
be3f702
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
server.httproutes are served at/agents/{deployment}/{path}through the LiveKit server; no local listener exposes user routes (a health-only listener remains for probes)publicviaconfigure_endpoint) registers on the control connection; the worker dials a fixed pool of wires speakingAgentHttp.Framedraining/seqfor server-side load balancingBuilt on #6856 (base branch). Depends on livekit/protocol#1728 — requires a livekit-protocol Python release with the new bindings, so CI fails until that ships and the pin is raised.