Skip to content

Qredence/fleet-pi

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

217 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Fleet Pi

Fleet Pi

Latest release License CI Node pnpm

A local browser workspace for Pi-powered coding agents — with durable memory, resumable plans, and repo-scoped tools you can review in Git.

Quick start · Features · Docs · Releases


Most agent tools hide memory, plans, and session state in the cloud or in logs you cannot diff. Fleet Pi runs on your machine, keeps adaptive state in agent-workspace/, and gives you a full chat UI with Agent and Plan modes before anything mutates your repo.

Why Fleet Pi?

Git-native state Memory, plans, skills, and artifacts live in reviewable files under agent-workspace/
Plan before you ship Plan mode inspects the repo read-only, asks follow-up questions, and produces numbered plans you can execute later
Your credentials Default model is Google Gemini (gemini-3.5-flash); Bedrock and other Pi providers still work
Your Pi stack Project skills, prompts, and extensions load from .pi/ and agent-workspace/pi/

Quick start

Prerequisites: Node.js 22+, pnpm 10.33.3 (via Corepack), and LLM credentials through Pi's normal provider auth flow.

git clone https://github.com/Qredence/fleet-pi.git
cd fleet-pi
corepack enable
pnpm install
cp .env.example .env
pnpm dev

Open http://localhost:3000, then sanity-check the server:

curl http://localhost:3000/api/health
# → {"status":"ok"}

In the chat UI, try read package.json and confirm a Read tool card appears.

Need more detail? See docs/quickstart.md for standalone vs Codex paths, provider setup, and smoke checks.

Features

Chat Persistent Pi sessions, streaming NDJSON, resume after refresh
Agent mode Repo-scoped read, write, edit, bash plus approved Pi extensions
Plan mode Read-only exploration, structured plans, InputBar questions, execute / refine / stay
OpenUI Inline generative UI blocks inside assistant messages
Resources Browse skills, prompts, extensions, and workspace files from the side panel
Memory Durable project memory and recall under agent-workspace/memory/
Web access web_search, fetch_content, and code_search in Agent mode (pi-web-access)
Optional Neon mirror Mirror Pi sessions to Postgres when FLEET_PI_CHAT_DATABASE_URL is set — JSONL stays source of truth

Built with TanStack Start, React 19, Pi coding agent, and @workspace/hax-design (agent-elements + OpenUI).

How it works

Browser chat UI  →  /api/chat  →  Pi AgentSession  →  repo-scoped tools
                      ↓
              agent-workspace/  +  .pi/settings.json  +  .fleet/sessions/
  1. You send a message from the web app.
  2. Fleet Pi resumes or creates a Pi session scoped to the active project root.
  3. Tool calls run on the server inside that boundary; Plan mode blocks mutating commands.
  4. Durable context is written to workspace files you can commit, diff, and share.

Setup paths

Standalone (recommended)

Run Fleet Pi as a normal local web app. No Codex desktop app required — just Node, pnpm, and working LLM provider credentials.

Pi / Codex

Use the shared Codex environment and worktree bootstrap when you want Fleet Pi inside a Codex workflow:

Optional integrations

Set these in .env only when you need them — local dev works without them.

Integration Env vars What it enables
Neon session mirror FLEET_PI_CHAT_DATABASE_URL Queryable mirror of Pi sessions, runs, and tool events
App auth BETTER_AUTH_SECRET, FLEET_PI_AUTH_DATABASE_URL Multi-user login (email/password + optional Google OAuth)
Daytona sandboxes DAYTONA_API_KEY Isolated container execution for file/bash tools
Amazon Bedrock AWS_REGION, AWS_PROFILE Use Bedrock models instead of the Gemini default

See .env.example for the full list.

Key concepts

Agent vs Plan mode

  • Agent — full coding tools and approved external Pi tools.
  • Plan — read-only inspection, follow-up questions, numbered Plan: steps, and resumable plan cards after refresh.

agent-workspace/

agent-workspace/ is Fleet Pi's durable adaptive layer: memory, plans, evals, artifacts, and chat-installed Pi resources. Human docs live in docs/; agent-facing files live here. See docs/agent-workspace.md.

Project-local Pi resources

Committed resources under .pi/ surface in the resources browser. Chat-driven installs land in agent-workspace/pi/ and stay reviewable in Git.

Docs

Document Description
docs/README.md Docs hub and reading order
docs/quickstart.md Setup, providers, and verification
docs/agent-workspace.md Durable workspace model
docs/codex.md Advanced Codex / worktree path
docs/architecture.md Generated architecture reference
docs/api.md Generated API reference
RELEASE_NOTES_v0.5.0.md Latest release highlights

Development

pnpm lint
pnpm typecheck
pnpm test
pnpm build

Extra checks: pnpm e2e, pnpm syncpack, pnpm generate:docs, pnpm validate-agents-md.

Path Role
apps/web/ TanStack Start app, /api/chat, and file routes
packages/hax-design/ Shared UI — agent-elements, OpenUI, Fleet Pi surfaces
agent-workspace/ Durable agent memory and workspace artifacts
.pi/ Pi settings, skills, prompts, and extensions

apps/web/src/routeTree.gen.ts is generated — do not edit by hand.

Contributing

Contributions welcome. Please read CONTRIBUTING.md before opening a PR.

License

Fleet Pi is released under the Apache 2.0 License.

About

Fleet Pi , workspace for Pi-powered coding agents, with durable plans, memory, and repo-scoped tools

Topics

Resources

License

Code of conduct

Contributing

Security policy

Stars

Watchers

Forks

Packages

 
 
 

Contributors