OpenHands / Agent Canvas

The current OpenHands repository presents Agent Canvas, a self-hosted control center for coding agents and automations. It runs the OpenHands agent and can connect other supported agents, including Claude Code and Codex. Backends can run locally or on remote infrastructure. Current project overview.

It is useful when you want one place to manage agent conversations and repeated development work across projects. A team must still choose the agent, execution environment, credentials, and checks for each task.

Set up a bounded repository task

Start with the repository’s current quickstart and select the backend that matches your environment. Limit the available project directories and configure the agent’s model access. Ask for a small change or an explanation of a failing test. Inspect commands, file changes, and the resulting checks before accepting the work.

Execution isolation depends on the selected backend and configuration. The repository notes that isolating conversation execution does not place the entire Canvas or automation service in a sandbox. Review the documented setup before giving it sensitive project access.

Free software does not cover every service

The repository has an MIT license. The current pricing page lists Agent Canvas as free to self-host with your own model access. It describes a single user or shared identity for this option.

Teams is a separate paid self-hosted offering with multiple user identities and administrative controls. Enterprise has custom pricing. Model-provider usage, third-party agent subscriptions, machines, and any cloud services have their own costs. An MIT license does not include free inference or unlimited hosted execution.

Checked against official sources on October 23, 2026. This listing reflects the current Agent Canvas presentation and does not report an installation or performance test.