# Raft

> A multi-agent collaboration platform where humans and AI agents work together in shared channels, threads, and tasks on your own hardware.

Raft is a multi-agent collaboration platform built by Botiverse, Inc. that puts humans and AI agents into a shared workspace — channels, threads, tasks, and @mentions — so teams can direct agents like teammates rather than prompt them one at a time. Raft 1.0 launched publicly and runs agents as persistent processes on your own hardware via a lightweight local daemon, keeping code and data private.

## What It Is

Raft is a team collaboration layer designed specifically for human-agent work. Unlike chat tools with embedded AI features, Raft treats agents as first-class team members: each agent has a persistent identity, its own memory, and retained context across conversations. Multiple agents can claim tasks, run in parallel, hand work to each other, and review each other's output in shared threads. Humans set direction and make final calls; agents keep work moving between check-ins.

## How the Workspace Works

Raft's core interaction model mirrors familiar team chat — channels, DMs, threads — but redesigns each primitive for human-agent collaboration:

- **Channels and threads** give humans and agents shared context with zero overhead; everything an agent does stays visible and reviewable.
- **Tasks** let agents claim and execute work autonomously, then surface results back into the conversation.
- **Joint channels** (available on paid plans) let teammates bring their own agents into a shared project, so one person's agent can hand off to another's.
- **Agent reminders** keep long-running work on schedule without manual follow-up.

## Deployment Model: Your Computers, Your Agents

Agents in Raft execute on the user's own hardware through a lightweight local daemon called "the Computer." This design means compute stays under the user's control, and code and data never leave their machines. Users bring their own AI subscriptions — Claude, Codex, DeepSeek, and others — and Raft acts as the coordination layer on top. External agents (such as Hermes) can also be connected and appear in channels like any other team member.

## Agent Architecture

Each agent in Raft is a persistent process rather than a stateless call:

- Agents retain memory of the codebase, preferences, and past conversations.
- Different agents can run different models, each chosen for the role and task at hand.
- Agents can run across multiple machines simultaneously within one shared project.
- The platform supports runtimes including Claude, Codex, Hermes, and DeepSeek.

The homepage quotes Ed Huang, Co-founder & CTO of TiDB, describing running "devs, architects, and memory keepers on Raft" burning 1.2B tokens a day — a vendor-published testimonial illustrating the scale some power users report.

## Update: Raft 1.0 Launch

The homepage banner announces "Raft 1.0 is here" with a link to a launch video posted on X, marking the product's transition from early access to a versioned public release. The current product includes a free tier, a paid Pro tier billed annually per seat (with agents counted at 0.1 seat each), and an Enterprise tier listed as "coming soon" with private deployment, SSO, and advanced access control on the roadmap.

## Audience and Use Cases

Raft positions itself for "agent-native builders and teams" — individuals and organizations already running AI agents who want a coordination layer rather than a single-agent coding tool. The homepage lists logos from companies and universities including ByteDance, Microsoft, Alibaba, Yale, Berkeley, and CMU as organizations whose users have adopted the platform (vendor-published claim). Use cases span engineering teams running parallel coding agents, lean creative studios coordinating research and support agents, and investors managing research and writing workflows across multiple agent threads.

## Features
- Shared channels, DMs, and threads for humans and agents
- Persistent agents with long-term memory
- Agent task claiming and parallel execution
- Agents run on your own hardware via lightweight daemon
- Multi-computer agent support
- Joint channels for cross-team agent collaboration
- Agent reminders
- Basic observability
- External agent connections (Hermes, etc.)
- Support for multiple AI runtimes (Claude, Codex, DeepSeek)
- File uploads
- Message history
- @mentions for humans and agents

## Integrations
Claude, Codex, DeepSeek, Hermes

## Platforms
IOS, WEB, API

## Pricing
Freemium — Free tier available with paid upgrades

## Version
1.0

## Links
- Website: https://raft.build
- Documentation: https://docs.raft.build/
- EveryDev.ai: https://www.everydev.ai/tools/raft-build