Tabby
by TabbyML
Self-hosted, open-source AI coding assistant for privacy-conscious teams
Last reviewed 2026-06-19
Tabby is an open-source, self-hosted AI coding assistant from TabbyML, positioned as a privacy-first alternative to cloud copilots like GitHub Copilot. It runs on your own hardware (including consumer GPUs) with no required database or cloud service, providing repo-aware code completion, chat about code, user-invoked inline edits, and an Answer Engine that does retrieval Q&A over indexed repositories and docs with a visible source/thinking trace. Tabby is model-agnostic (it works with open models such as CodeLlama, StarCoder, Qwen, and DeepSeek) and exposes an OpenAI-compatible API. The core is Apache 2.0 licensed, with enterprise features under a separate license. It is aimed at developers and teams who need on-prem or air-gapped AI coding for security, compliance, or proprietary-code reasons. An agentic "Agent" mode is in private preview rather than generally available.
What it can do
Complete code (repo-aware)
CopilotProvides multi-line, full-function completions aware of the repository; the developer accepts or rejects each suggestion.
sourceChat about code
AssistantAnswers questions and explains code via an in-editor chat ("Explain This").
sourceApply user-invoked inline edits
CopilotPerforms single edits on demand (Ctrl/Cmd+I) at the developer's request; not autonomous.
sourceAnswer questions over indexed repos and docs (Answer Engine)
AssistantDoes retrieval Q&A over indexed Git repos and custom docs with a visible source and thinking trace and shareable Pages.
source
Strengths
- +True self-hosted/on-prem, no cloud or database dependency; runs on consumer GPUs for air-gapped or regulated environments
- +Open source (Apache 2.0 core), model-agnostic, with an OpenAI-compatible API
- +Actively maintained with broad IDE coverage, repo-aware completion, and an Answer Engine
Limitations
- −Capability ceiling versus cloud incumbents: fundamentally completion, chat, and retrieval
- −Agentic workflows and MCP are roadmap or private preview, not generally available
- −Self-hosting carries operational burden, and quality depends on the open models you serve
Overview
Tabby is an open-source, self-hosted AI coding assistant from TabbyML, positioned as a privacy-first alternative to cloud copilots. It runs on your own hardware with no required database or cloud service.
What it does
Tabby provides repo-aware code completion, in-editor chat about code, user-invoked inline edits, and an Answer Engine that does retrieval Q&A over indexed repos and docs with a visible source/thinking trace and shareable Pages. It is model-agnostic (CodeLlama, StarCoder, Qwen, DeepSeek and others) and exposes an OpenAI-compatible API. An agentic "Agent" mode is in private preview.
Integrations & setup
Works with VS Code, JetBrains, and Vim/Neovim, plus GitHub and GitLab connectors. Ships as a Docker image and runs on consumer GPUs. Access is via REST/OpenAI-compatible endpoints; MCP is roadmap-only.
Pricing
The Apache 2.0 core is free (Community). Team and Enterprise tiers add flexible deployment and enhanced security/support and are contact-sales; the ee/ enterprise directory is under a separate license.
Best for / not for
Best for developers and teams that need self-hosted or air-gapped AI coding for privacy, security, or compliance. Less suited to teams that want frontier-grade agentic workflows out of the box.
Traction
Tabby raised a reported $3.2M seed announced in October 2023 (Yunqi Partners, ZooCap) and has a large, active open-source community.
Alternatives
GitHub Copilot is the cloud incumbent; Continue and Sourcegraph Cody are other open/self-hostable options.
What people are saying
We aggregate real LinkedIn discussion into sentiment for the agents people search most. Tabby isn't tracked yet, want it added? Request tracking.
FAQ
Is Tabby an autonomous coding agent?+
No. Tabby is primarily a copilot/assistant: repo-aware completion, chat, user-invoked inline edits, and an Answer Engine. An agentic "Agent" mode is in private preview, not generally available, so it should not be treated as an autonomous agent.
Can Tabby run fully on-premise?+
Yes. It is designed to be self-hosted with no required database or cloud service and can run on consumer GPUs, which is its main appeal for privacy- and compliance-sensitive teams.
Sources
- Tabby (official site) · accessed 2026-06-19
- Tabby documentation · accessed 2026-06-19
- TabbyML/tabby (GitHub) · accessed 2026-06-19
Last reviewed 2026-06-19