Tabby

Open-source, self-hosted AI coding assistant

Freemium· Community (self-hosted) free; Team and Enterprise tiers with paid licensing (contact sales for pricing).CodingDeepSeek-Coder, Qwen2.5-Coder, StarCoder, CodeLlama, or any OpenAI-compatible endpoint (Mistral, etc.)

Visit website →

Best for

Engineering teams in regulated or security-sensitive environments that need an in-IDE coding assistant and codebase Q&A running entirely on their own infrastructure.

Skip if

Solo developers who just want the fastest possible autocomplete with zero setup, or teams that need cutting-edge agentic multi-file editing and are fine sending code to a SaaS.

Tabby (TabbyML) is an open-source, self-hosted AI coding assistant designed as a privacy-preserving alternative to cloud copilots like GitHub Copilot, Cursor, or Sourcegraph Cody. It bundles the pieces a development team needs to run its own coding AI: a completion engine, an in-editor chat, an 'Answer Engine' that lets developers ask questions against their internal codebase and connected knowledge sources, and connectors that pull context from Git repositories, issue trackers, and docs. The whole stack runs behind your firewall on your own hardware, so source code never leaves the network. Tabby is model-agnostic — it ships with support for open weights such as DeepSeek-Coder, Qwen2.5-Coder, StarCoder, and CodeLlama, and can also proxy to hosted providers like Mistral, OpenAI-compatible endpoints, and others. It runs on modest consumer-grade GPUs (a single 24GB card is enough for most teams) and provides IDE plugins for VS Code, the JetBrains family, Neovim, Eclipse, and Android Studio. Typical workflows include ghost-text code completion tuned on the team's own repos, retrieval-augmented chat that cites internal files, PR/commit summarization, and using the Answer Engine as an internal StackOverflow that indexes wikis and Git history. Admin features include SSO, role-based access, usage analytics, and an OpenAPI-documented HTTP API for scripting and integrations. It fits regulated industries (finance, defense, healthcare) and any engineering org that cannot ship proprietary code to a third-party inference API.

Editor's take

Tabby is the most credible self-hosted answer to Copilot right now. If your legal team has ever red-lined a cloud AI vendor, this is the tool to pilot — you trade a bit of raw model quality for full control, and the Answer Engine over your own repos is genuinely useful once it's indexed.

— The AI Tool Bible editorial team

Pros

✅ Fully open source (Apache 2.0) and self-hostable — no code leaves your infrastructure
✅ Model-agnostic: swap between open-weight coders (DeepSeek, Qwen2.5-Coder, StarCoder) or route to hosted APIs
✅ Answer Engine grounds chat in the team's own repos, issues, and docs via built-in connectors
✅ Runs on a single consumer GPU; lightweight enough for on-prem or air-gapped deployment
✅ Wide IDE coverage — VS Code, all JetBrains IDEs, Neovim, Eclipse, Android Studio
✅ Documented HTTP/OpenAPI surface for custom integrations and CI tooling
✅ Enterprise features (SSO, RBAC, analytics) available without leaving the self-hosted model

Cons

⚠️ Completion quality with open-weight models still trails frontier hosted assistants like Copilot or Cursor on complex refactors
⚠️ Requires GPU capacity, ops effort, and someone to maintain the deployment — not turnkey
⚠️ Smaller plugin ecosystem and fewer agentic/multi-file editing features than commercial competitors
⚠️ Team/Enterprise pricing is not published; requires a sales conversation
⚠️ Retrieval quality depends on how well you configure connectors and index your repos

Use cases

Self-hosted code completionIn-IDE AI chatCodebase question answeringInternal knowledge search over repos and docsAir-gapped AI coding for regulated industriesPR and commit summarizationOnboarding assistance for new engineersCustom model deployment for coding