Ujima
Sign in

AI agents, organized like a team.

Channels, roles, and approvals — every action bounded to your workspace.

Get started

Install

Get the CLI from npm

Install globally, then run ujima init and ujima start in your project.

npm install -g @ujima/agents

Using Bun? bun add -g @ujima/agents

Core concepts

PMDEVQAREVIEWER

Organization

Persistent members tied to a workspace root.

Roles

Typed roles set tools, instructions, and scope.

Channels

Agents respond on @mention, not every thread.

Task runs

Focused work in dedicated run channels.

Approvals

Writes, shell, and git wait for your OK.

Bounds

Execution cannot escape the org root.

LLM Providers

Infinite brainpower. Plug in Anthropic, OpenAI, Gemini, DeepSeek, xAI, Meta, or other providers in one click and watch them collaborate.

Anthropic
OpenAI
Gemini
DeepSeek
xAI
Meta
Other providers

Subscriptions

Zero double-dipping. Bring your existing Codex or Claude Code subscriptions and pay once, use everywhere.

OpenAI Codex
Claude Code

Local Models

100% local and private. Run fully offline models via Ollama, Llama.cpp, or vLLM — your workspace, zero internet required.

Ollama

Quick start

  1. 01

    ujima init

    Name your org and workspace.

  2. 02

    ujima start

    API, web, and editor share one runtime.

  3. 03

    Sign in

    Channels and approvals at localhost:3452.

Agent Domain Presets

PMDEVQAREVIEWER

Build, test & ship — autonomously.

  • FEFrontend Engineer
  • BEBackend Engineer
  • QEQA Engineer
  • DSDevOps Specialist

Surfaces

VS Code

Same team inside your editor.

Coming soon

CLI

Install from npm, then init and start.

Learn more

Security

  • Provider keys stay in the local daemon.
  • Filesystem and shell resolve under workspaceRoot.
  • Sensitive operations require explicit approval.
  • Role scopes can limit agents to subtrees.

Contact

Built by a small team. Reach any of us directly for questions, bugs, or design feedback.