OpenHands

You install Agent Canvas and run it locally or on a shared VM, then connect a model with your own API key and point it at a repository. From there, work arrives rather than being typed: you define an automation that watches for a signal, such as a new pull request, a failed CI workflow, an @mention in a Slack channel, a new Linear issue or a nightly schedule. When the signal fires, the agent opens a conversation with that context attached, gathers what it needs, plans the work, executes it, validates the result and opens a pull request. You watch each step and keep the merge decision. Integrations reach the rest of the stack through MCP, so internal tools, APIs and data sources are available to the agent without bespoke glue. Templates cover the common cases, including PR review, issue triage, CI failure repair, security remediation and incident retrospectives, and the Agent Control Plane and SDK exist for teams wiring it into something larger.

• •Bring Your Own Agent - Runs Claude Code, Codex, Gemini CLI, OpenHands or your own stack, so the platform and the agent are separate choices
• •Model and Runtime Agnostic - Closed models or open weights including Kimi, DeepSeek and Qwen, executing locally, in a cloud sandbox or inside your VPC
• •Event-Triggered Automations - Workflows that start from GitHub, Slack, Jira, Linear, a push or a schedule rather than from somebody opening a tab
• •The Full Loop - Signal, plan, execute, validate, ship and monitor, with humans keeping review, approval and merge
• •MCP Integrations - Internal tools, APIs and data sources connected through the open Model Context Protocol
• •MIT Licensed Core - Agent Canvas is free to self-host with no seat fee, and the paid tiers add identity, RBAC, SSO and budget controls rather than capability

Teams that already have more repetitive engineering work than people to do it: pull request review, CI failures, dependency bumps, issue triage, QA sweeps and release chores, all of which are high volume, low judgment, and well suited to something that starts without being asked. It is the clearest fit for organisations that want this on their own infrastructure and their own model keys, and for anyone unwilling to commit a whole workflow to a single agent vendor.

Be clear about what it is not. The free tier is a single shared identity, so a team that needs per-user accounts is looking at $500 a month, and there is nothing between those two points: a solo developer or a pair has no middle option. The homepage now leads with enterprise workflow marketing (Slack triggers, Jira incident routing, retrospective drafting) rather than the terminal coding agent many people still arrive expecting, and OpenHands Cloud publishes no usage or credit price at all, so the hosted product's cost is only visible after you sign in. If you want a coding assistant inside your editor, this is a heavier thing than you need.