Your code
could use a hand.

Hands is an open source platform for software engineering agents. Start them from one canvas, run them on your machine, your servers or a cloud, and let them work while you do something else.

Get started Docs GitHub

Introducing Hands

The new agent avatars.

Twelve hands on one plane, each with its own colour and its own cycle of moves. An idea for how a team of agents could look when they all work at once.

Every agent in a conversation gets a little hand of its own: a colour, a mood that follows what it is doing, and a motion for waiting, thinking and finishing. They are made with Kosmos, the same characters and engine work behind Kamiji.

Git, in 3D

  • committing
  • opening a PR
  • merged
  • checks passed
  • checks failed
  • resolving a conflict

Works with the Kamiji libraries

A Hands avatar is a Kamiji character, so everything in the Kamiji libraries is compatible: any character, face, emoji, gesture or icon can be an agent's avatar or accessory, with the same colours, outlines and shadows.

  • Kamikin mochi

    14 characters

    Any of the 14 Kamikin characters can be an agent's avatar, with six faces and nine clips.

    See the library
  • Rubbernose

    6 characters

    1930s cartoon things with faces (an eraser, a clock, a soda can and more) as an agent's avatar, with six faces, nine clips and a tintable body.

    See the library
  • Hand poses

    13 poses

    Thumbs up, OK, victory, wave, clap, pray and more, as an agent's gesture for a state.

    See the library
  • Emoji faces

    33 faces

    Round emoji heads with their own expressions, for agents that should look like a face.

    See the library
  • Emotes and reactions

    24 emotes, 6 reactions

    Hearts, sparkles, zzz and more, popped in around an avatar or tapped as reactions.

    See the library
  • Symbols and objects

    36 symbols, 34 objects

    Status and tool symbols and everyday objects, usable as badges and accessories.

    See the library
  • Body shapes

    36 shapes

    Spheres, beans, drops and character heads, tweened from the base mesh.

    See the library
  • Outlines, toon and shadow

    options

    Ink outline, banded toon shading and a soft ground shadow on any of the above.

    See the library

Counts are the files in the Kamiji repository today. The libraries are free for non-commercial use under CC BY-NC 4.0; for-profit use needs the Premium Licence (draft). See the licence section.

What Hands is

Hands brings coding agents together in one place. You start a conversation, point it at a project, and an agent reads the code, runs commands, edits files and reports back. You can also set agents to run on a schedule or when something happens, such as a GitHub event.

It is a control center rather than a single model. It runs its own open source agent out of the box, can drive other agents that speak the Agent Client Protocol, and works with the language model of your choice.

Products

  • Agent Canvas GUI

    The browser control center for agent conversations and automations. It runs locally by default and can switch between several backends. The README marks it as beta.

  • Agent Server server

    A REST and WebSocket server that runs agents on a host. Run it on a laptop, a dedicated machine, a virtual machine or in the cloud, and connect the canvas to as many as you like.

  • Software Agent SDK SDK

    A Python library for building agents that use tools, workspaces and conversations. Use it to put agents inside your own tools.

  • CLI terminal

    The same agent in your terminal, installed with uv or an install script.

  • Automation Server schedules

    Starts agents on a schedule or from webhook events, and keeps a run history. It is a separate repository that pairs with the Agent Server.

  • Cloud and Enterprise hosted

    A managed cloud with hosted execution, integrations and usage controls, and an enterprise option for licensed self-hosting or managed deployment.

Sources: Agent Canvas README, docs.openhands.dev, openhands.dev, github.com/OpenHands.

How it works

  • Agents

    The built-in OpenHands agent, plus Claude Code, Codex, Gemini CLI, Pi, OpenCode or any agent that speaks the Agent Client Protocol (ACP).

  • Runtimes

    Run on the host directly, in one Docker sandbox, or in a separate container for every conversation. The SDK also documents Apptainer and API-based sandboxes.

  • Backends

    Connect the canvas to local, remote and cloud agent servers and switch between them without leaving the page. Share one server with a team and keep a personal one on your laptop.

  • Subagents

    An agent can hand work to other agents. The SDK has a task tool set and file-based agent definitions for this.

  • Automations

    Reports to Slack, GitHub issues broken into tasks, and similar jobs, run on a schedule or from a webhook.

  • MCP and skills

    Connect tools through the Model Context Protocol and give agents reusable skills. Public skills live in the extensions repository.

  • Any model

    Bring your own language model and keep several LLM profiles.

Get started

Check the sources. Package and image names still say openhands. Commands are copied from the files named under each block.

Agent Canvas from the repo

Needs Node.js 24 or later and uv. This runs the agent server on your machine with access to your files, so read the README warning first.

npm install -g @openhands/agent-canvas
agent-canvas

Source: README.md, Option 1. Then open http://localhost:8000.

Docker sandbox

export PROJECTS_PATH="$HOME/projects"
mkdir -p "$PROJECTS_PATH" "$HOME/.openhands"

docker run -it --rm \
  -p 127.0.0.1:8000:8000 \
  -e AGENT_CANVAS_ALLOW_LAN_SESSION_KEY=true \
  -v "$HOME/.openhands:/home/openhands/.openhands" \
  -v "${PROJECTS_PATH}:/projects" \
  ghcr.io/openhands/agent-canvas:1.29.0

Source: README.md, Option 2. Docker serves the UI at http://localhost:8000/canvas.

One container per conversation

npm install -g @openhands/agent-canvas
OH_CONVERSATION_RUNTIME=docker agent-canvas

Source: README.md, Option 3.

From source

git clone https://github.com/OpenHands/OpenHands.git
cd OpenHands
npm install
npm run dev

Source: README.md, Option 4.

CLI and SDK from the docs

CLI

uv tool install openhands --python 3.12
openhands

Source: docs.openhands.dev, CLI installation page, fetched 24 Oct 2026. Needs Python 3.12 or later.

Python SDK

pip install -U openhands-sdk openhands-tools

Source: docs.openhands.dev/sdk/getting-started. Install both packages together so versions match.

Integrations

  • GitHub documented

    Cloud and Enterprise installation guides exist, and automations can react to GitHub events.

  • GitLab documented

    Cloud and Enterprise installation guides exist.

  • Slack documented

    Cloud and Enterprise guides exist, and automations can publish reports to Slack.

  • Linear and Notion automations

    Named in the Agent Canvas README as services that automations can connect to.

  • MCP servers documented

    Add tools through the Model Context Protocol in settings or in the SDK.

Meet the agents

An idea, not a shipped feature: each agent in a conversation gets its own little character as a profile picture, so you can tell them apart at a glance. These are live Kamiji models. The names are examples.

  • PlannerKamiji hand
  • ReviewerKamikin Hana
  • FixerKamikin Kon
  • ScribeKamikin Pip

Open source and licence

  • Hands code: MIT. The Agent Canvas repository carries the MIT licence, and GitHub lists MIT for the SDK, automation and extensions repositories too.
  • Kamiji avatar assets on this page (the hand and the Kamikin mochi) are not part of the Hands licence. They are CC BY-NC 4.0 with a paid Premium Licence for commercial use. See LICENSE.md in the Kamiji repository. Credit: "Kamiji by Paul Bloch".
  • The page shows no OpenHands logos or third-party marks.

Community and links