Tabby
Self-hosted AI coding assistant you can run on your own hardware
About
Tabby is an open-source, self-hosted AI coding assistant for teams that cannot send code to a vendor. Rather than a plugin that phones home, it is a server you deploy yourself — on your own machine, your own cloud or an air-gapped network — and it provides real-time code completion, an answer engine for questions about your codebase, a code browser and pluggable context providers. IDE and editor clients connect to that server. The pricing fits that model. Community is $0 and supports up to five users with unlimited free code completions, which Tabby guarantees is never limited or feature-gated — the only time they may ask for a card is to prevent abuse from extreme usage. Team is $19 per user per month for up to 50 users and adds usage reports, analytics and the ability to enforce IDE and extension telemetry policies. Enterprise is quote-only, removes the user cap and adds an authentication domain, SSO and a dedicated Slack channel with roadmap priority. Tabby Cloud, for teams that want the same product without running it, bills on model token cost with $20 of free usage each month, auto-charges once usage passes $10 and lets each user set a budget cap. The trade-offs are the classic self-hosting ones: you own uptime, upgrades and model hosting, and completion quality depends on the model you choose to serve. The ecosystem is far smaller than Copilot's, and the tool is best understood as a privacy and data-residency play rather than a feature race. One detail worth noticing: the usage-billing copy on the pricing page still refers to 'Pochi' in one paragraph, a sign the vendor is consolidating products — confirm current terms before a large commitment.
Key Features
- Self-hosted server with real-time code completion
- Answer engine over your own codebase
- Code browser and pluggable context providers
- Usage reports, analytics and telemetry policy enforcement
- SSO and authentication domain on Enterprise
- Optional cloud version billed on model token cost
Deals, Discounts & How to Save
Tabby is cheapest when you self-host and serve your own model: the Community edition is free for up to five users with unlimited completions, and there is no token bill at all because inference happens on your hardware. Above five users the Team tier at $19/user/month is still well below mainstream assistants, and the cloud version never charges without a budget cap you set.
Pros
- Runs entirely on your own infrastructure, including air-gapped networks
- Community edition is free for up to 5 users with unlimited completions
- Team tier can enforce IDE and extension telemetry policies
- No per-token billing when you host the model yourself
Cons
- You own uptime, upgrades and model serving
- Smaller ecosystem than mainstream commercial assistants
- Enterprise terms and pricing are quote-only
- Completion quality depends entirely on the model you run
Pro Tips for Tabby
Start with the free Community edition for up to five users — completions are never gated.
Price the hardware before you commit: self-hosting moves cost from licences to GPU time.
On the cloud version, set per-user budget caps before onboarding the team.
Use the Team tier's telemetry policy enforcement if you need to prove which extensions are active.
Ask about the product consolidation hinted at in the pricing copy before signing a long enterprise deal.
Alternatives to Tabby
GitHub Copilot
FeaturedAI pair programmer in every IDE
Tabnine
Enterprise-grade AI code assistant with air-gapped deployment
Continue
Open-source AI code assistant for any IDE
Refact.ai
Local-first agentic coding engine with a free tier and self-hosting
More in Ai Code Editor
Cursor
FeaturedAI-first code editor built on VS Code
GitHub Copilot
FeaturedAI pair programmer in every IDE
Windsurf
FeaturedAgentic IDE (formerly Windsurf), now sold as Devin Desktop
Cody
Sourcegraph's AI coding assistant