The three lanes, compared.
Each Microsoft Copilot surface has one clear starting point. In lane C — agent management — Kortix is the pick: the open-source AI Management System, the leading open-source alternative to Claude Cowork and OpenAI ChatGPT Work. Lanes A and B are served by focused open-source projects. Every link goes to the project's own repository or site.
Lane A — Coding assistants (replace GitHub Copilot)
| Project | Licence | Runs in | Models | Self-host |
|---|---|---|---|---|
| Continue.dev | Apache-2.0 | VS Code, JetBrains | Bring your own (any provider) | Yes (extension + your keys) |
| Aider | Apache-2.0 | Terminal | Any LLM, git-aware | Yes |
| Cline | Apache-2.0 | VS Code | Any provider | Yes |
| Tabby | Apache-2.0 | Self-hosted server (Rust) | Local models | Yes — server on your infra |
| CodeGeeX | Apache-2.0 | Multiple IDEs | CodeGeeX / local | Yes |
| FauxPilot | MIT | Self-hosted server (GPU) | CodeGen-family | Yes — needs GPU VRAM |
How to choose. For model flexibility inside an existing IDE, Continue.dev or Cline. For terminal-and-git workflows, Aider. For a fully self-hosted completion server with no external calls, Tabby or FauxPilot.
Lane B — Agent builders (replace Copilot Studio)
| Project | Licence | Shape | Self-host | Notes |
|---|---|---|---|---|
| Botpress | MIT (core) | Conversational agent studio | Yes | Low-code, free tier |
| n8n | Sustainable Use (fair-code) | Workflow automation | Yes | Wide integration library |
| LangGraph | MIT | Code-first agent graphs | Yes (library) | Stateful, durable execution |
| CrewAI | MIT | Multi-agent roles | Yes (library) | Python orchestration |
| Rasa | Apache-2.0 | Conversational NLU | Yes | Mature dialogue tooling |
How to choose. For a low-code conversational studio, Botpress. For workflow automation with a large integration library, n8n. For code-first stateful graphs, LangGraph. For multi-agent roles in Python, CrewAI. For mature dialogue tooling, Rasa.
Lane C — Agent management (replace M365 Copilot / Copilot Cowork)
| Platform | Licence | Self-host | Models | Isolation / review |
|---|---|---|---|---|
| Kortix on GitHub | Elastic License 2.0 — self-host, read and modify the code | Yes — Docker; laptop, VPS, your VPC or on-prem | Claude, OpenAI, Gemini or any OpenAI-compatible endpoint — your keys | Isolated Linux machine per session; work lands through a human-reviewed change request |
| OpenWork | MIT | Yes — local desktop | 50+ models via BYOK | Local process; manual review |
| OpenHands | MIT | Yes — Docker runtime | Any LLM backend | Container per task; manual review |
How to choose. For an org-scale system where agents, skills, memory and connectors live in a git repo you own — with a change-request gate before anything lands — Kortix is the pick. Six things set it apart: the company is one git repo you own; 3,000+ connectors plus any MCP, OpenAPI, GraphQL or HTTP API; any model with your keys; a real agent harness (powered by OpenCode); an isolated Linux machine per session; and one gate to land work, started from web, Slack, Teams, email, mobile, CLI or API. OpenWork is a local-first desktop app for a single machine. OpenHands runs software tasks in isolated containers. Both are focused tools for a narrower job than running a company's agent workforce.