Official descriptionOpen source8d ago
Qredence/fleet-pi
Fleet Pi , workspace for Pi-powered coding agents, with durable plans, memory, and repo-scoped tools
View on GitHub ↗Direct link · no redirect page
- Star count
- 6 ★ · Qredence/fleet-pi · read from the repository page · (checked 2026-10-16 20:00:20)
- Source
- read directly from its GitHub page · type: open-source project
- Harvest history
- first seen 2026-10-09 · last seen 2026-10-16 · seen 12× · listed by 1 sources
- Score
- score 103 · 6 ★ · 1 sources · repo
- Install
npm install cp
FAQ (answers come from our records)
How do I install it?
The repository provides this install command:
npm install cp. It comes from the repository itself; we cannot guarantee it runs as-is in your client.Who is it for?
We filed it under these tags: 开源项目. The tags come from the entry itself, not a taxonomy we imposed.
Who maintains it?
repository
Qredence/fleet-pi；first catalogued 2026-10-09；last seen 2026-10-16。There is no official certification or quality rating here — we record sources and dates only.Third-party quality signals
Fleet Pi
以下摘录自该仓库的 README（片段） · 在 GitHub 读完整 README →
**A self-improving agent workspace. You bring the keys and the sandbox. The agent builds the rest.**
Quick start · Why · Docs · Changelog · Pitches
---
Fleet Pi is a **meta-harness** — a complete, self-hosted agent workspace built on Pi. It ships as a working web app with a chat UI, Agent and Plan modes, durable workspace memory, and sandboxed execution. But unlike a traditional agent tool, Fleet Pi is designed to be **self-improving**: the agent writes its own skills, registers its own prompts, evaluates its own output, and adapts to you over time.
The agent is simultaneously the worker and the architect of its own environment. Two users with Fleet Pi will have completely different systems after a few sessions — because the system _became_ what they needed it to be.
agent-workspace/ is Fleet Pi's durable adaptive layer: its memory, plans, skills, evals, and artifacts stay with the project as reviewable files.
Why Fleet Pi?
| Problem | Fleet Pi's approach |
| ---------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------- |
| **SaaS lock-in** — your agent lives in their cloud, your data is proprietary | **Self-hosted, BYOK, BYO sandbox** — your keys, your compute, your Git repo |
| **Fixed feature set** — you're stuck on the vendor's roadmap | **Self-improving** — the agent creates its own skills, prompts, and tools |
| **No audit trail** — opaque logs you can't diff or review | **Git-native provenance** — every file change is reviewable, every decision is traceable |
| **No isolation** — shared tenants, no per-user sandboxing | **Per-user Daytona sandboxes** with durable volumes and credential isolation |
| **DIY complexity** — frameworks that give you parts, not a working system | **Ship as a working app** — clone, install, run, and the agent adapts |
Quick start
**Prerequisites:** Node.js 22+, pnpm (via Corepack), and an LLM provider key.
Open http://localhost:3000. Type read package.json to verify the agent is working.
**Need more detail?** See docs/quickstart.md.
Features
| Feature | Description |
| --------------------- | -------------------------------------------------------------------------------- |
| **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 |
| **Harness mode** | Evaluation-oriented tools with workspace writes and Autocontext status |
| **OpenUI** | Inline generative UI blocks inside assistant messages |
| **Resources browser** | Browse skills, prompts, extensions, and workspace files from the side panel |
| **Workspace memory** | Durable project memory under agent-workspace/memory/ |
| **Web access** | web_search, fetch_content, code_search in Agent mode |
| **Subagents** | Scout (fast recon), researcher (web research), worker (code changes) |
| **Extensions** | Full TypeScript extension system; 5,300+ community Pi packages |
Quick start · Why · Docs · Changelog · Pitches
---
Fleet Pi is a **meta-harness** — a complete, self-hosted agent workspace built on Pi. It ships as a working web app with a chat UI, Agent and Plan modes, durable workspace memory, and sandboxed execution. But unlike a traditional agent tool, Fleet Pi is designed to be **self-improving**: the agent writes its own skills, registers its own prompts, evaluates its own output, and adapts to you over time.
The agent is simultaneously the worker and the architect of its own environment. Two users with Fleet Pi will have completely different systems after a few sessions — because the system _became_ what they needed it to be.
agent-workspace/ is Fleet Pi's durable adaptive layer: its memory, plans, skills, evals, and artifacts stay with the project as reviewable files.
Why Fleet Pi?
| Problem | Fleet Pi's approach |
| ---------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------- |
| **SaaS lock-in** — your agent lives in their cloud, your data is proprietary | **Self-hosted, BYOK, BYO sandbox** — your keys, your compute, your Git repo |
| **Fixed feature set** — you're stuck on the vendor's roadmap | **Self-improving** — the agent creates its own skills, prompts, and tools |
| **No audit trail** — opaque logs you can't diff or review | **Git-native provenance** — every file change is reviewable, every decision is traceable |
| **No isolation** — shared tenants, no per-user sandboxing | **Per-user Daytona sandboxes** with durable volumes and credential isolation |
| **DIY complexity** — frameworks that give you parts, not a working system | **Ship as a working app** — clone, install, run, and the agent adapts |
Quick start
**Prerequisites:** Node.js 22+, pnpm (via Corepack), and an LLM provider key.
Open http://localhost:3000. Type read package.json to verify the agent is working.
**Need more detail?** See docs/quickstart.md.
Features
| Feature | Description |
| --------------------- | -------------------------------------------------------------------------------- |
| **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 |
| **Harness mode** | Evaluation-oriented tools with workspace writes and Autocontext status |
| **OpenUI** | Inline generative UI blocks inside assistant messages |
| **Resources browser** | Browse skills, prompts, extensions, and workspace files from the side panel |
| **Workspace memory** | Durable project memory under agent-workspace/memory/ |
| **Web access** | web_search, fetch_content, code_search in Agent mode |
| **Subagents** | Scout (fast recon), researcher (web research), worker (code changes) |
| **Extensions** | Full TypeScript extension system; 5,300+ community Pi packages |
- Source
- github · https://github.com/Qredence/fleet-pi
- Browse by field
- Open source
- Traction
- 6 ★
- added
- yesterday
More in Open source
Filed under
Every tag opens the full list for that filter. Counts are the live catalogue figures.