Features Quick Start Docs About Discord GitHub Get Started →
Free & Open Source

OtterBot

A free, open-source, local-first multi-agent AI workspace. Build a roster of AI agents — each with its own personality, memory, skills, and model — and watch them coordinate. Runs entirely on your machine.

otterbot — bash
____ _ _ ____ _
/ __ \| | | | | _ \ | |
| | | | |_| |_ ___ _ _| |_) | ___ | |_
| | | | __| __/ _ \ '__| _ < / _ \| __|
| |__| | |_| || __/ | | |_) | (_) | |_
\____/ \__|\__\___|_| |____/ \___/ \__|
 
$ npm install -g otterbot
Installed otterbot
$ otterbot start
Generated OTTERBOT_DB_KEY
Daemon listening on port 3001
$ otterbot web
Opening http://localhost:3001
$

A team of agents, on your machine

Multi-Agent Roster

Build a roster of independent agents. Each one has its own personality, model, memory, skills, and credentials — not modes of a single bot.

Agent Studio

Tune every agent in one place: identity, persona, model, skills, schedule, memory, and credentials — each on its own tab.

Memory & Skills

Every agent remembers across sessions with hybrid keyword and vector search, and loads markdown skill packs scoped just to it.

Live Message Bus

Agents coordinate over a message bus you can watch in real time. A COO agent leads the team, and trusted agents spawn subagents for focused work.

Local-First & Encrypted

Runs on the machine you already have. Every database is encrypted at rest — your conversations and credentials never leave home.

Bring Your Own Model

Point each agent at Anthropic, OpenAI, LM Studio, or Ollama. Mix fast local models and frontier cloud models across your roster.

Up and running in three commands

# Install the CLI, start the daemon, open the web UI
$ npm install -g otterbot
$ otterbot start
$ otterbot web
🦦

What happens on first run? otterbot start launches a background daemon and generates an encryption key for your local databases. otterbot web opens the UI, where a setup wizard configures your first agent — the COO. All data stays in ~/.otterbot. See the docs for CLI commands and configuration.

About

Built for the rest of us

After using OpenClaw for a few days, I was frustrated. I could never tell what was actually happening. Everything had to be built from scratch with zero guidance. Cost-efficient models delivered a terrible experience, and the higher-tier models had response times that made conversations painful.

When I talk to an assistant, I want an immediate response. Assistants don’t need to be brilliant — they just need to coordinate the right agents. OtterBot lets you point each agent at whatever model fits: a fast local model on LM Studio or Ollama for the coordinator, a frontier model from Anthropic or OpenAI for the heavy lifting. No burning through expensive API credits just to hold a conversation.

Combine that with the live message bus and 3D view, and you always know exactly what your agents are doing.

Get Started Read the Docs