← all SaaS
Self-host GitHub Copilot
AI code completion / chat in the IDE ·
Category: AI / LLM tooling
GitHub Copilot is a paid IDE plugin that streams code completions and chat answers from a hosted model. The self-hostable replacements run an open-weight code model (DeepSeek-Coder, Qwen2.5-Coder, StarCoder2) on your own GPU and serve it through an editor extension that mimics Copilot's tab-complete + chat UX.
GitHub Copilot pricing anchor: Copilot Individual $10/user/mo; Business $19/user/mo; Enterprise $39/user/mo.
- GitHub
- ★ 33.1k · last commit 3w ago · 714 open issues
- License
-
Apache-2.0 - Setup time
- 10min — VS Code or JetBrains plugin + a local model
- Monthly cost
- Free for the plugin; model server is the cost line — $0 on a workstation 16GB+ GPU, ~$200/mo on cloud GPU.
Migration sketch. Install the Continue extension from the VS Code / JetBrains marketplace. Edit `~/.continue/config.json` to add an Ollama provider with `model: qwen2.5-coder:7b` for completions and a larger model (Llama 3.1 70B / DeepSeek-Coder-V2 33B) for chat. Uninstall Copilot from the same IDE — they fight over the inline-suggestion API. Continue's autocomplete latency on a 7B model is comparable to Copilot at <300ms on a recent GPU.
Good fit forDevelopers who already run a local model and want the Copilot UX (ghost-text + chat + edit) without the hosted bill.
Weak atMobile / web IDEs (Codespaces, GitHub.dev) — Continue is desktop-IDE only.
- GitHub
- ★ 33.5k · last commit 2mo ago · 313 open issues
- License
-
Apache-2.0 Core is Apache-2.0; some enterprise extras (SAML, SSO controls) are gated.
- Setup time
- 20min docker run with --gpus + plugin install
- Monthly cost
- $200+/mo cloud GPU for the inference server; $0 on a workstation. Plugin overhead is negligible.
Migration sketch. Run `docker run --gpus all -p 8080:8080 -v $HOME/.tabby:/data tabbyml/tabby serve --model StarCoder-1B --device cuda`. Install the Tabby IDE extension (VS Code / JetBrains / Vim / Emacs) and point it at `http://your-host:8080`. Tabby is purpose-built as a Copilot-replacement server — it has a multi-user admin UI, repo-aware indexing, and per-team usage analytics that Continue doesn't try to match.
Good fit forTeams that want one centrally-hosted code-AI server with admin controls and per-developer usage tracking.
Weak atChat / agentic edits — Tabby's strength is completion; for chat-heavy workflows pair with Continue or Aider.
- GitHub
- ★ 44.6k · last commit 2w ago · 1531 open issues
- License
-
Apache-2.0 - Setup time
- 5min `pip install aider-chat`
- Monthly cost
- Free CLI; model cost is the line item — pair with Ollama / vLLM for $0 marginal cost.
Migration sketch. `pip install aider-chat`, then run `aider --model ollama_chat/qwen2.5-coder:32b` inside any git repo. Aider operates at the git-diff level — it reads files you /add, proposes edits, and commits them. Replaces Copilot Chat's edit-mode and Cursor's composer rather than inline ghost-text completion.
Good fit forTerminal-first developers who want a pair-programmer that operates on whole files and commits diffs to git.
Weak atInline tab-complete during typing — Aider is a chat/edit loop, not a ghost-text engine; pair with Continue if you want both.
In a terminal? npx os-alt github-copilot prints this table —
how the CLI works →