Fleet Prime is a local workspace for coding and research with AI. Open a project in your browser, chat with an agent, and follow its work as it reads files, runs commands, and produces results. Fleet provides the interface and runs the stock, Rust-based Prime Agent 0.10+ runtime. The exact version is pinned in PRIME_AGENT_RUNTIME.json.
fleet-prime-0.5.8.mp4
- Work across projects and return to persistent sessions from the sidebar.
- Watch responses and tool activity as the agent works, including commands, file edits, plans, questions, and subagents.
- Browse workspace files, attach material to a conversation, and inspect generated artifacts.
- Use a managed IPython environment for interactive analysis and file operations.
- Configure providers and models, and use planning and refinement flows in the web interface.
- macOS or Linux on an x64 or ARM64 machine. Windows is not supported.
- Node.js 22.12.0 or later.
- Python 3.10 or later for the managed IPython kernel.
Install Fleet globally:
npm install --global @qredence/fleetStart Fleet from the project directory you want to work in:
cd /path/to/your/project
fleet-agentOpen the local URL printed by the launcher. On first use, add a provider in
Settings → Providers or run /login, then choose a model in the composer.
Start a conversation by describing what you want to work on. You can return to
your sessions and switch projects from the sidebar.
fleet-prime is an alias for fleet-agent. To use the upstream terminal
interface instead, run:
fleet-agent agentOn first launch, Fleet downloads its pinned Prime Agent runtime, verifies its
checksums, and caches it under ~/.prime/fleet/runtimes. A separately installed
prime-agent CLI does not change Fleet's pinned runtime. The pin is recorded in
PRIME_AGENT_RUNTIME.json.
Fleet supports existing Prime Agent settings and session data from the earlier TypeScript-based runtime through a legacy compatibility layer. It runs the pinned Rust runtime and does not attach to an older TypeScript daemon.
If Fleet reports that its daemon socket is owned by an incompatible daemon,
stop the existing daemon and restart Fleet. Fleet uses a per-user socket under
$TMPDIR/prime-agent-<uid>; if socket creation fails because the path is too
long, use a shorter TMPDIR such as /tmp. An advanced installation can set
FLEET_PRIME_AGENT_BIN to use a different native binary.
Fleet runs locally and the web launcher accepts loopback connections only. The agent can run model-generated Python and project commands with your user permissions; Fleet is not a security sandbox. Use trusted repositories, instructions, skills, and extensions, and review changes before accepting them. See SECURITY.md for vulnerability reporting.
To develop Fleet, clone the repository, install its workspace dependencies with pnpm, and start the web application:
git clone https://github.com/Qredence/fleet-prime-agent.git
cd fleet-prime-agent
pnpm install
pnpm run dev:webSee CONTRIBUTING.md for requirements, checks, and the pull request process. The documentation index links to the architecture, adapter contract, and operation-specific guides.
Thanks to everyone who has contributed to Fleet:
This list is generated from merged pull requests: each release pull request adds anyone new, and each GitHub release credits the contributors of that release. The upstream engine's contributors are credited in Prime Agent.
- Ask questions or share ideas on GitHub Discussions.
- Report bugs through the repository's issue tracker.
- Read SUPPORT.md for community and support options, or the Prime Agent documentation for the upstream engine.
Fleet Prime Agent is released under the MIT License. Fleet is powered by Prime Agent, whose lineage includes pi-mono by Mario Zechner.