Helix
A self-hosted coding agent platform runs parallel tasks in isolated desktops, with human review and local inference through Ollama or vLLM.

Helix is a paid, self-hosted platform for engineering teams coordinating coding agents. Agents work on separate tasks in parallel, while the team shares a view of their progress and reviews their work.
- Isolated desktops: Each task gets a Linux desktop with an editor, browser and terminal, plus separate filesystems, credentials and network access.
- Specifications and review: A shared Kanban board tracks work through planning and implementation, with specification approval before coding and human sign-off before pull requests merge in GitHub, GitLab or Azure DevOps.
- Shared supervision: Teammates can watch desktops in their browsers, inspect files, use terminals and take the keyboard; chat history and running environments persist across handoffs.
- Agent and model choice: Use Claude Code, OpenAI Codex, Gemini CLI, Qwen Code, Goose, Zed Agent or other ACP-compatible agents, and switch agents or models mid-task while retaining the conversation and uncommitted work.
- Code context and team controls: Kodit indexes repositories and generates code wikis, while role-based access, SSO, audit logs and usage tracking support shared projects.
Helix runs on macOS, Linux GPU systems or Kubernetes, and also supports Docker deployment. Local inference works with Ollama, LM Studio, vLLM and OpenAI-compatible endpoints. It supports air-gapped operation with local models; the Mac offering describes fully offline use on machines with 64 GB or more of unified memory.
The Mac offering keeps credentials and development data on your hardware. Hosted model providers receive inference requests when you choose them, through API keys or supported Claude and ChatGPT subscriptions. Managed Helix Cloud runs the agent infrastructure on hosted servers.
The wider platform includes document retrieval, vision RAG, MCP tools, scheduled tasks and webhook triggers.