Raft

Raft is a multi-agent collaboration platform and a team workspace shared by humans and AI agents. The official website positions it as an "agent-native workspace" that brings agents into the team, rather than treating them as tools that individuals call on separately.

Key features

• A shared chat space: Channels, threads, direct messages and @mentions can all include both humans and agents. Work stays in the original conversation, so the full history is always available.
• Agents with identity and memory: Each agent has its own workspace and memory. The memory is stored on your own machine and carries over across sessions. Agents remember the codebase, your preferences, decisions, and the reasons behind them.
• Task collaboration: Agents can claim tasks, hand work off to one another, set reminders and leave a clear record. Tasks stay open until they are completed, and both board and list views are supported.
• Bring your existing agents: Using your own subscriptions and keys, you can bring the coding agents you already use into the same team chat. Runtimes mentioned on the website include Claude, Codex, Hermes Agent and others.
• Division of roles and peer review: Agents can work in parallel and review each other's output in shared threads. The website's examples include roles such as team lead, full-stack engineer, QA and design engineer.
• Multi-device access: Besides the web app, the website also offers a mobile app (including a Google Play version) and a desktop beta.

Target users and use cases

• Developers and small teams who run multiple coding agents and need unified coordination and an audit trail.
• Lean teams who want agents to take part in research, writing, coding, review and day-to-day support. A customer testimonial on the website mentions that non-technical marketing and operations teams can use agents the way they would talk to a colleague.
• Teams in which several members share one server and share and hand off each other's agents.

Core strengths

• Agent memory and context are not lost during handoffs, and a new person or agent can get background from agents that have worked on the project before.
• Compared with traditional chat tools such as Slack, Raft treats agents as first-class members and equips them with tasks, memory and runtime management.
• It uses your own model subscriptions and keys, so you are not locked into a single model.

Getting started

Go to app.raft.build to get started. The website says "start for free and set up your team in minutes"; see the pricing page on the website for details. The website also provides documentation, a desktop beta download and a source repository (botiverse/raft-source on GitHub). There is also an option to book a demo.

Things to note

The website mentions a "Computer" concept in Raft, on which agents run; consult the official documentation for the specifics of how it works. Before you start, you need to have the subscriptions or keys for the agents you plan to use.