Skip to content
Zachary BENSALEM edited this page May 22, 2026 · 2 revisions

Fleet Pi

Fleet Pi is a browser-based coding-agent workspace built on Qredence's Pi AI coding agent. It pairs a streaming chat UI with a persistent agent session, letting you have a full conversation with an AI that can read, write, and run code inside your repository. A durable agent-workspace/ directory accumulates skills, memory, plans, and artifacts across sessions.

What it does

  • Streams responses from Pi, backed by Amazon Bedrock (Claude models)
  • Runs Pi's built-in tools — read, write, edit, bash, grep, find, ls — scoped to your project root
  • Supports Agent mode (full tools) and Plan mode (read-only planning with numbered steps)
  • Maintains persistent Pi JSONL sessions; restores transcript on page refresh
  • Optionally mirrors sessions to Neon Postgres, provisions Daytona sandboxes per user, and renders inline generative UI via OpenUI

Quick links

Topic Page
Getting started Getting started
Architecture overview Architecture
Chat API endpoints API
Plan mode Plan mode
Agent workspace Agent workspace
Configuration reference Configuration

Repository layout

fleet-pi/
├── apps/web/          TanStack Start app (React 19 + Vite + Nitro)
├── packages/ui/       Shared component library (@workspace/ui)
├── agent-workspace/   Durable agent home (skills, memory, plans)
├── .pi/               Pi runtime configuration and extensions
└── scripts/           Tooling scripts (auth migration, doc generation)

Generated by Factory

Clone this wiki locally