Tabby AI coding assistant for self-hosted development teams
Tabby is a self-hosted AI coding assistant. Learn what it does, where businesses use it and how MILabs sets it up.
Updated 2026-08-03
What Tabby is
Tabby is an open-source AI coding assistant from TabbyML that developers use inside their code editors. It suggests code as a person types, answers questions about a codebase and works through the editor extensions that Tabby publishes for common development environments. What sets it apart is the deployment model. Tabby is built to be self-hosted, which means the server that runs the model can sit on hardware a business owns or rents, rather than on a shared service run by a third party. For a company that wants source code to stay within its own infrastructure, that choice matters. Tabby suits teams whose engineers already write code daily and who have a reason to keep code and prompts in-house. The model and hardware choices are settled during setup. Tabby works best as one part of a wider development process, alongside code review, testing and the judgement of engineers.
Tabby in plain language
An AI coding assistant reads the code around the cursor and suggests the next few lines, so a developer spends less time on repetitive typing and more on design decisions. Some assistants work through a service run by the vendor, where code is sent to the vendor's servers for each request. A self-hosted assistant takes the opposite approach. The model and the serving software run on a machine the company controls, whether that is a server in its own office, a rented machine in a data centre or a private cloud account it manages.
The trade-off is that someone has to look after that machine. Someone chooses the model, watches memory and processing load, applies updates and decides who may connect. Those tasks are modest for a team with an internal IT function, but they are real. Capability also depends on the hardware, so a larger model needs more capacity than a small one. Teams usually weigh these costs against the value of keeping source code in place and having a clear record of who can use the service.
What businesses use it for
- Helping engineers complete routine functions and tests while they work in their editor
- Answering questions about an existing codebase so new staff can find their way
- Keeping source code and prompts within infrastructure the business controls
- Drafting comments for internal modules that have little documentation
- Explaining unfamiliar code during review or handover
Who it suits
- Software teams that need code assistance on infrastructure they control
- Engineering leads who want clear ownership of an internal developer tool
- Businesses with sensitive source code and an in-house IT function
How MILabs sets it up
MILabs starts by agreeing with the owner or team lead which deployment option and configuration suit the business, including where the server will run and who may use it. Next, MILabs connects Tabby to the systems already in place, such as source repositories, the editors engineers use and any internal documentation the assistant should draw on. Permissions follow, so access to the server, the model settings and the repositories it can read is limited to the people who need it, and that list is written down. Finally, the business agrees who looks after Tabby, covering updates, monitoring and questions from staff. MILabs works remotely from Sydney and runs setup sessions by video call, with decisions recorded as they are made.
Builds it fits
Industries it fits
Tabby in practice
In a typical engineering team the assistant sits quietly in the editor for most of the day. A developer writing a data import script sees suggested loops and error handling, accepts the parts that fit and rewrites the rest. Developers who maintain older services often use chat-style questions to trace where a function is called from, which saves a search through several repositories. Such use tends to be ad hoc rather than scheduled, so the team gains most from agreeing a few shared habits early.
A lead might ask each engineer to note which suggestions were useful and which needed correction, then review those notes at the regular team meeting. Code review still belongs to people. Reviewers look for logic errors that a suggestion can carry into the codebase, and they check that generated tests actually assert something meaningful. The server side needs a named person as well, who watches usage, applies model updates in a test environment first and keeps the list of accounts current when someone leaves the team.
Frequently asked questions
- Can Tabby connect to the software I already use?
- Yes, most directly through the code editors. Tabby works inside those editors through its extensions, so its closest links are the editor your engineers already use and the code repositories it can read. Connections to other business software, such as ticketing or document stores, usually need a separate build. MILabs can design that link so each system keeps a clear role within the workflow.
- How does Tabby compare with other tools for the same job?
- Many hosted assistants, such as GitHub Copilot or Cursor, rely on the vendor's own service. Tabby's distinguishing feature is that it is built to run on a server the business controls. The better choice depends on whether code location, hardware cost or governance matters more than convenience. MILabs can help compare the options on those points before a decision is made.
- How do I get my team to use Tabby?
- Start with a small group of engineers who already want better code suggestions, and let them set up the editor extension first. Share a short written guide covering what the assistant is for and how its answers fit into code review. Gather feedback during real work and adjust the setup. Adoption tends to follow when people see it help with tasks they already do.
- Can I self-host Tabby?
- Yes. Tabby is designed to be self-hosted, so the server that runs the model can operate on hardware the business owns or rents. Hardware needs depend on the model and the number of people using it, and MILabs works through those needs with the owner. Where the server runs and who can reach it are planned early and written down with the owner.
- Which builds work well with Tabby?
- Builds that answer questions from internal material suit a self-hosted setup best, such as the staff knowledge assistant, the document question-and-answer assistant and the private document server. A knowledge base chatbot works well when the same material also needs to reach customers. MILabs chooses the build after learning what the team needs to find and who will use it.
