Tabby
Tabby is an open-source, self-hosted AI coding assistant that aims to enhance and streamline the coding experience. With Tabby, users can e...
Last verified:
What is Tabby?
Tabby is an open-source, self-hosted AI coding assistant that runs your code-completion and conversational coding workflows from infrastructure you control. It provides an on-premises alternative to cloud-hosted copilots by offering a local inference server, repository indexing (RAG) for context-aware completions, and integrations that let the assistant operate inside editors and other developer tools. Key features include fast Fill-In-The-Middle (FIM) code completion optimized for developer workflows, a chat-based assistant for repository questions and debugging, and support for running local or GPU-backed models with an OpenAPI-compatible interface. Tabby is aimed at engineering teams, enterprises, and privacy-conscious developers who need private, customizable AI tooling integrated into IDEs and CI environments.
Tabby pricing
Pricing model: Free
Tabby itself is distributed as open-source software and can be self-hosted for free; there is no on-site licensing fee for running the project code. Costs come from infrastructure (VMs/GPUs/storage), model licensing (if using proprietary models), and optional hosted LLM API usage (e.g., OpenAI) if you choose to proxy inference rather than run models locally. The website and docs describe free-to-run source code and community releases, while paid or hosted services (if any) would be separate and depend on third-party model or cloud provider fees.
Tabby pros
- Self-hosted deployment keeps code and telemetry private
- Open-source codebase allows inspection and customization
- Repository indexing (RAG) provides code-aware answers
- Fill-In-The-Middle (FIM) optimized for code completion
- Supports running models on consumer and datacenter GPUs
- OpenAPI-compatible API for easy integration
- Local inference reduces dependency on external cloud services
- IDE integrations (VSCode/JetBrains) embed assistance in editor
- Can operate with multiple models (conversation + coding)
- Offers a chat interface for repository questions and debugging
- No required DBMS or external cloud to get started
- Configurable to use hosted LLM APIs for low-latency setups
- Active GitHub repository and community for issues and contributions
- Command-line tooling for model download and server management
- Designed for teams to run private assistants on their infra
Tabby cons
- Requires infrastructure and ops knowledge to self-host
- Model downloads can be large and require significant disk space
- Setting up GPU acceleration requires compatible drivers and hardware
- Offline installation and model setup can be non-trivial
- Some features may require external models or API keys
- Performance depends on chosen model and local hardware
- Limited turnkey cloud-hosted offering compared to managed copilots
- Troubleshooting model compatibility may require reading docs or issues
Frequently asked questions about Tabby
Can I run Tabby fully offline on my local network?
Yes — Tabby is designed to be self-hosted and can run on local infrastructure, but fully offline setups require downloading compatible model files and embeddings ahead of time and configuring the local registry and model paths; some community reports and documentation note additional steps for offline model setup.
Which editors does Tabby integrate with?
Tabby provides editor extensions and integrations for common developer IDEs such as VSCode and JetBrains family editors, enabling in-editor code completion, chat, and other assistant features once the local Tabby server is running and connected.
What models can Tabby run?
Tabby supports running a variety of open and commercial models that are compatible with its inference server; you can run smaller consumer-grade or larger datacenter models on CUDA-backed GPUs or configure Tabby to use external hosted APIs for inference depending on latency and hardware constraints.
How does Tabby use repository context for answers?
Tabby indexes repositories to build retrieval-augmented generation (RAG) context so that code completions and chat responses can reference the project’s codebase, which improves relevance for debugging, code navigation, and context-aware completions.
Is Tabby free to use for teams?
The Tabby software is open-source and freely available to self-host; teams incur operational costs for the infrastructure (compute, GPUs, storage) and any third-party model or API fees if they choose to use paid hosted models rather than local model binaries.
What are the hardware requirements to run Tabby well?
Hardware requirements depend on the model size you choose: consumer-grade GPUs can run smaller models for interactive completion, while larger models require datacenter-class GPUs and more memory; the docs recommend matching model choice to available GPU and CPU resources and provide guidance for CUDA device usage.
How do I update Tabby and its models?
Tabby is updated via its releases and GitHub repository; models are managed separately (downloaded and placed in model directories or fetched via provided tooling), so updating involves pulling newer Tabby releases and updating or re-downloading model files as needed.
Does Tabby collect telemetry or user data by default?
Because Tabby is self-hosted and open-source, telemetry behavior is under user control; deploying on your own infrastructure means you decide what telemetry or logging is enabled and where logs are stored, aligning with privacy needs.
Can Tabby use hosted LLMs instead of local models?
Yes — Tabby can be configured to proxy or call hosted LLM APIs for inference to achieve lower latency or avoid local GPU provisioning, though this introduces external API costs and requires managing API keys.
Where can I get support or report bugs?
Support and issue tracking happen through the project’s GitHub repository and community channels (such as Slack and discussion forums) where maintainers and contributors triage issues, feature requests, and bugs.