Sourcegraph Cody vs Tabnine
Tabnine is an AI code assistant with chat, autocomplete, and slash commands. It was originally a plugin focused on autocomplete, but it has more recently expanded to offer more features similar to other AI code assistants. Tabnine is unique because it uses a combination of local, on-device LLMs and cloud-based LLMs. The local LLM is used to provide "basic code completions" (short completions of 2-3 words) while the cloud models power longer completions and chat.
Tabnine's Pro and Enterprise tiers give users access to their propriety models and third-party models from Anthropic and OpenAI. However, Tabnine only promises security features such as zero retention for its proprietary models. They say that retention with other models, such as those from OpenAI and Mistral, is determined by the respective vendors' public terms of service. Tabnine offers fully private deployments that use their proprietary models.
Lastly, Tabnine does use local code context to inform the LLM, but remote code context (such as for code living on GitHub and GitLab) is currently in private preview for their Enterprise tier.
TL;DR: Tabnine uses a unique blend of local and cloud-based models, and is a good choice for individuals who want a free and purely local AI code assistant. Cody is a better choice for free users who want access to more powerful models or for enterprises that want a solution that has context awareness for a full enterprise codebase.