OpenHands Agent Canvas: A Self-Hostable AI Coding Platform

This article introduces OpenHands Agent Canvas—a self-hostable open-source AI coding platform. It emphasizes local-first and autonomous control, and can be deployed locally, on a remote VM, in a private cloud, or within an enterprise intranet. It supports three execution modes: manual trigger, scheduled run, and event-driven. It is compatible with various agents like Claude Code and Codex, as well as mainstream LLMs. Currently in the sandbox stage and incubated by the community, its goal is to make AI coding assistants truly belong to developers themselves.

OpenHands Agent Canvas addresses the lack of autonomous control in traditional cloud-based AI coding tools by proposing a local-first solution. Incubated by the OpenHands community, the project is in the sandbox stage—its code may not have been fully tested—but its vision is to let developers have full control over their AI coding assistants. Official note: This project may be AI-assisted generated, not fully tested, or outdated, and no responsibility is taken for the code or support.

Core Features: Supports three operation modes: 1. Manual trigger (direct prompt input interaction); 2. Scheduled run (integrate with Automation Server to set up scheduled tasks like code review and document updates); 3. Event-driven (listen to external events like GitHub PRs, Slack mentions, and Webhooks for automatic execution).

Deployment Options: 1. Local development environment (Node.js 22.12+, installed and run via npm); 2. Remote virtual machine (provides SELF_HOSTING.md guide); 3. Cloud hosting (OpenHands Cloud commercial service); 4. Enterprise intranet (integration with internal tools).

Agent Canvas is not tied to specific agents or models:

• Supported agent types: Claude Code, Codex, custom agents (complying with the OpenHands Agent Server protocol);
• Supported LLM providers: Anthropic (Claude series), OpenAI (GPT series), Google (Gemini series), Mistral, Minimax, Kimi (MoonShot AI). Developers can choose the appropriate model based on their needs to avoid vendor lock-in.

Technical Architecture: Frontend-backend separation. The backend core is the Agent Server (manages agent lifecycle, file interaction, sandbox environment, log collection); optional Automation Server (supports scheduled/event tasks); frontend includes browser panel, chat panel, file browser, and settings interface.

Security Considerations: Default Docker deployment isolates agents; containers run with host UID/GID to limit file access; the dockerless mode (dev:dangerously-dockerless) directly exposes the host file system and should be used with caution.

• Individual developers: Daily coding assistant (generate unit tests, refactor code, write comments);
• Small teams: Shared server deployment for unified code style and automated maintenance;
• Open-source projects: GitHub integration for automatic PR initial screening and Issue responses;
• Enterprise environments: Private cloud deployment to meet compliance requirements, integrate with internal systems, and enable fine-grained access control.

OpenHands Agent Canvas represents the direction of decentralized AI coding tools, empowering developers with full control over their AI assistants (hardware, models, rules, data). Although in the sandbox stage (unstable API, incomplete features), the community is actively iterating. For developers focused on autonomous control, it is an open-source project worth attention, and is expected to become an important infrastructure for AI-assisted programming in the future.