Skip to main content
Fleet Prime Agent is a coding agent you run locally. It keeps a persistent Python kernel per session, spawns real subagents through rlm(...), streams every turn as structured events, and improves itself over time from your feedback with /refine. You can drive it from four surfaces that share the same runtime:
  • CLI — one-shot or scripted runs (text, json, rpc modes).
  • TUI — full interactive terminal.
  • Daemon — long-lived background process that CLI, TUI, and web attach to.
  • Web app — the Qredence UI for chat, tool cards, session branching.

​
Start here

Quickstart

Install, sign in a provider, and run your first agent turn in under five minutes.

Install

System requirements and install steps for the agent workspace and web UI.

Core concepts

Sessions, subagents, streaming events, refinement — the ideas the rest of the docs assume.

Feature guides

Task-oriented walkthroughs for chat, tools, providers, sessions, MCP, and OAuth.

​
What you can do with it

Chat with a model that can run code

Every session gets its own persistent IPython kernel plus shell, edit, grep, and find tools.

Choose from 19 providers

OpenAI, Anthropic, Bedrock, Gemini, and 15 more, with per-model cost and thinking-level metadata.

Branch and resume sessions

Fork any point in a transcript. Compaction keeps long runs inside the model’s context window.

Spawn subagents from Python

Call rlm(...) inside an IPython cell to run a real child agent in parallel or in the background.

Add tools with MCP or extensions

Register external tool servers over MCP, or ship your own extension package.

Teach it with skills

Turn a recurring workflow into a personal or project skill loaded automatically.
Last modified on August 17, 2026