How Tabnine Protects Your Code Privacy

TL;DR:

Your code always remains private

Tabnine NEVER stores or shares any of your code. Any action that shares your code with the Tabnine servers for the purpose of training team models requires explicit opt-in. Tabnine does not retain any user code beyond the immediate time frame required for training models. Any team model created by Tabnine is only accessible by your team members.

Trained on open-source code with permissive licenses

Tabnine only uses open-source code with permissive licenses for our Public Code trained AI model (MIT, Apache 2.0, BSD-2-Clause, BSD-3-Clause). Whether you’re using Tabnine’s Pro plan or our Basic plan, your code, and AI data are NEVER used to train any models other than your own team models

Tabnine Privacy Deep Dive

What is Tabnine?

Tabnine provides AI code completions using two sources: (1) A Public Code model that has been trained on open-source code. (2) A Team Learning model trained on you and your team’s code and local interactions with Tabnine.

How do we build our models?

Training the Open-Source model

The Open-Source model is trained on public code from GitHub.
It only uses repositories with permissive open-source licenses (MIT, Apache 2.0, BSD-2-Clause, BSD-3-Clause).
Trained code is filtered to ensure quality and avoid outdated code, esoteric code, auto-generated code, and other edge cases.
The model is updated regularly to capture recent developments.

Training the Team AI model

Your private Team Learning model is trained on your code and interactions with Tabnine.
The model and AI training data are never shared with anyone outside your team.
The Team Learning model captures patterns that are specific to your team.
The Team Learning model requires an explicit opt-in from the team administrator.
Every Team Learning model is trained in a secure and separate environment.

How do we use our models?

Tabnine uses different models to deliver the best possible experience and quality.

Local mode

The model is downloaded to your machine, and the data is queried locally, enabling offline work. You receive code completions continuously and quickly in this mode, however, the models do run on local resources. Consequently, these models are also smaller and not as strong than other Tabnine models. For some setups, the local mode may be too heavy.

Cloud mode

Models that are larger, stronger, and more diverse will run on Tabnine Cloud on larger and faster machines. In cloud mode, the predictions are more accurate and longer, are lighter on the local machine, and require a good network in order to function.

For both models, your code and AI training data are only used to provide the service, and your code is never shared nor used as part of Tabnine’s open-source trained AI.

Due to their inherent benefits, as of June 1, 2022 both models will be enabled by default for all new installations, allowing for the best quality and smoothest user experience (you can opt-out of either model, at any time).

Tabnine for Teams

The Public Code AI model is downloaded from the Tabnine server.
Inference runs locally. Your code never leaves your machine, not even as context.
Inference uses two models: The Public Code model and Team Learning model.
The Team Learning model is trained continuously on your code and your team’s interactions with code completions.
The cumulative interactions are used to fine-tune The Team Learning model, which in turn improves code completions for the entire team by identifying the most relevant patterns.