metis. Explore GitHub
AN OPEN WORKSPACE FOR AI AGENTS

Wisdom,
put to work.

Your models. Your tools. A workspace that connects them.
Research, build and automate with agents that can use
a browser, work with files and remember the context.

Get started Discover Metis
OPEN SOURCE/SELF-HOSTED/YOURS
An imagined limestone sculpture of Metis, surrounded by metallic orbits and fragments of light.
ANCIENT WISDOM. NEW POSSIBILITIES.FIG. 01

MANY MINDS. ONE WORKSPACE.

Bring your favorites.

Choose the intelligence.
Connect the tools. Keep the context.

MODELS & AGENT RUNTIMES
TOOLS & MCP INTEGRATIONS

Connect with your own accounts and keys. Availability and capabilities depend on the provider or MCP server; some integrations require additional setup.

Explore connections ↗

01 / THE WORKSPACE

From a thought
to something real.

Conversations are just the beginning.
Give your agents the context and tools
to carry the work forward.

METIS AI / INTERACTIVE PRODUCT VIEWLOCAL MOCK DATA

Good afternoon, Admin

Context  0 / —Usage  85%

This is a fully local interface mockup based on the live Metis viewport. Try the sidebar, chats, menus, composer and workspace tools — no request leaves this page.

02 / BUILT TO DO MORE

Intelligence needs
room to act.

Read the documentation
I.

A browser that can act.

Give an agent a real browser to research pages, work through forms, upload files and download results. Persistent profiles keep sessions available across tasks, with tabs and screenshots inside your workspace.

PERSISTENT SESSIONS · BROWSER TOOLS
II.

A team for the task.

Let a lead agent delegate focused work to subagents, each with its own conversation and run. Explore a codebase, investigate alternatives or review a change in parallel, then bring the findings together.

SUBAGENTS · MODEL SELECTION
III.

Context, without the reset.

Give each project its own instructions, files and conversations. Save useful facts and preferences as memories, keep notes nearby and carry relevant context into the next task.

PROJECTS · MEMORY · NOTES
IV.

Work that keeps moving.

Long tasks run in background workers, so closing a browser tab does not end the job. Inspect progress, respond to questions and return to stored run history, with checkpoint recovery where supported.

BACKGROUND RUNS · RECOVERY
V.

Your recurring work, handled.

Schedule a task once or on a recurring cadence. Each automation keeps a conversation history and its own browser state. Use tool routines for defined sequences, with dry runs before execution.

SCHEDULES · TOOL ROUTINES
VI.

Thinking you can see.

Turn an explanation into a canvas with formatted documents, diagrams, equations and charts. Explore mathematical ideas with interactive graphs and sliders, all alongside the conversation.

CANVASES · DIAGRAMS · GRAPHS

THE DETAILS MAKE IT YOURS

One place.
A lot of possibility.

The everyday tools that turn an agent into a useful part of your workflow.

Files, terminals & remote machines01

Read and edit project files, run commands and work with Git through connected tools. Enroll a remote client to bring that machine’s terminal and files into Metis, with an editor in the workspace.

Skills, MCP & custom modes02

Add reusable instructions as skills and connect MCP servers for additional tools. Choose Agent, Plan or Ask mode, or define a custom mode with its own instructions and tool permissions.

Voice input & richer conversations03

Speak a prompt using recording and transcription options. Attach files, inspect streamed tool activity, pin important messages and search or archive conversations as your work grows.

Share the result04

Create a share link for a conversation, optionally protect it with a password, and revoke it later. Shared chats can include attachments and be cloned for a new line of work.

Your models, your infrastructure05

Run Metis on your own machine or server. Connect cloud APIs, local models through Ollama, compatible endpoints or supported agent runtimes. Configure users, model access and workspace permissions from the administration area.

THE SPACE BETWEEN AN IDEA AND ITS EXECUTION

Human intent.
Agent capability.

You bring the direction. Metis brings your models, context and tools together to help carry it out.

Make the connection

AN ANCIENT IDEA, REIMAGINED

The wisdom
to make things happen.

In Greek mythology, Metis embodies wisdom and good counsel. Her name speaks to a practical intelligence: knowing how to turn understanding into action.

That idea lives here. A space where your thinking meets the tools to do something with it.

The story behind the name

03 / WHAT WILL YOU MAKE OF IT?

A little direction.
A useful outcome.

Start with the work you already do.
Give it a workspace that can do more.

RESEARCH

From open tabs
to a clear picture.

Explore sources in the browser, collect the important details and turn the findings into a structured note or visual explanation.

“Compare these approaches and explain the trade-offs.”
BROWSER → NOTES → CANVAS
BUILD

From unfamiliar code
to a practical plan.

Give agents a project to explore. Delegate investigations, inspect files and use the terminal to carry the next change forward.

“Map this codebase and plan the next release.”
PROJECT → SUBAGENTS → TOOLS
AUTOMATE

From repeated effort
to a regular routine.

Schedule an agent to revisit a task. Keep its instructions, browser session and previous runs in one place so you can review the outcome.

“Check these sources each morning and summarize what changed.”
SCHEDULE → RUN → REVIEW

04 / MAKE IT YOURS

Your next idea
deserves a workspace.

Open source. Self-hosted. Ready for your way of working.

01Install Metis

Choose Docker or the installer for your platform.

02Connect a model

Add a provider account, API key or local endpoint.

03Give it a task

Create a project, add context and start a conversation.

curl -fsSL https://github.com/f1shyondrugs/metis-ai/releases/latest/download/metis-docker-install.sh -o metis-docker-install.sh
bash metis-docker-install.sh --version latest

Recommended for self-hosted deployments. Requires Docker.

Installation guide ↗View source on GitHub ↗

A FEW GOOD QUESTIONS

Before you
begin.

Read the docs ↗
What exactly is Metis?

Metis is an open-source, self-hosted AI agent workspace that you use in your browser. It brings conversations, models, tools, project context and background tasks into one place. You connect the providers and services you want to use.

Is Metis free to use?

The Metis source code is available under the MIT license. You provide the machine or server it runs on. Cloud models, subscriptions and connected services may have their own charges; Metis does not include those accounts or usage credits.

Where does my data go?

Metis stores workspace data on your installation. Prompts and task context are sent to the model providers and external tools you connect when they are used. For a local model setup, you can connect Ollama or a compatible local endpoint. Self-hosting the workspace does not automatically make every connected service local.

Can I use the models and subscriptions I already have?

Metis supports cloud APIs, local models and several agent runtimes, with API key, OAuth or local runtime connections depending on the provider. Supported models, tools and account requirements vary. Check the provider setup instructions for the connection you want to use.

Do tasks stop when I close the tab?

Agent runs are managed by background workers on your Metis installation. They can continue after you close the browser, as long as the host and required services remain running. Stored run history and checkpoint recovery help you return to the work; resume behavior depends on the provider.

How can I extend it?

Connect an MCP server for additional tools, add reusable skills or define custom modes. You can also enroll remote clients to work with other machines. The repository includes documentation for installation, providers, tools and the application architecture.