cell serve
Start HTTP API server for LLM commands
Synopsis
Starts an OpenAI-compatible HTTP server that proxies chat completions to LLM agent binaries (claude, opencode).
Endpoints:
POST /v1/chat/completions — OpenAI chat completions API
GET /api/v1/health — health check
The model field selects the agent: “claude”, “opencode”, or “claude/claude-sonnet-4-5” (agent/submodel).
Request:
{"model": "claude", "messages": [{"role": "user", "content": "explain this"}]}
Examples:
cell serve
cell serve --port 9090
cell serve [flags]
Options
-h, --help help for serve
--port int port to listen on (default 8484)
Options inherited from parent commands
--base-image string core image for scaffold Dockerfile (default: ghcr.io/dimmkirr/devcell:core-local)
--build rebuild image before running (forces --no-cache)
--debug plain-text mode plus stream full build log to stdout
--dry-run print docker run argv and exit without running
--engine string execution engine: docker or vagrant (default "docker")
--macos use macOS VM via Vagrant (alias for --engine=vagrant)
--plain-text disable spinners, use plain log output (for CI/non-TTY)
--session-name string session name for persistent home (~/.devcell/<name>)
--vagrant-box string Vagrant box name override
--vagrant-provider string Vagrant provider (e.g. utm) (default "utm")
SEE ALSO
- cell - Run AI coding agents in a devcell container