Skip to content

Coding Agent on cellp

cellp explores a private control plane for coding agents: build → POST /versionspreview with forked D1/KV/R2 → promote. We align with Cloudflare Agent Cloud (docs) — not a Cloudflare account on your metal.

Status: research and lab validation only. See Research for scope.

Lab targets (ordered)

These upstream projects are exercised on the local dev stack in contributor workflows. Order reflects dependency risk (Durable Objects, WebSockets, bundle size), not a public release date.

  1. cloudflare/agents-starter — partial: SPA and HTTP paths; Workers AI binding removed in lab overlay; chat inference needs your own model wiring on celld
  2. qaml-ai/pi-worker (examples/hello-agent) — supported in lab
  3. southpolesteve/opencode-do — partial: HTTP and SSE; model path uses placeholders unless you add providers

Next tier (research): fx on Workersfx.sh coding agent in a Worker (wasm + Durable Object session); needs gateway credentials for full model turns.

Agent & platform targets (research)

PlatformStatusNotes
MastraLab onlyLab fixture checks (agent, tools, workflows, D1/R2/memory on celld); bring your own LLM keys — not a product support tier
fx on WorkersLab only (partial)fx.sh wasm agent · not Agent Cloud
Cloudflare AgentsResearch / plannedAgent Cloud platform; lab path via agents-starter
Cloudflare OSResearch / plannedWorkspace north-star
Eve (Vercel)Research / plannedeve.dev/docs
PiResearch / plannedpi.dev · pi-agent-core in CF OS
Deep Agents (LangChain)Research / plannedLangGraph harness

Lab only = contributor lab fixture checks, not a hosting promise · Research / planned = not supported as product capability yet

Self-hosted Workers control plane. Not affiliated with Cloudflare or Vercel.