Supporting AI Features on Self Managed
75% of our customers are self managed, currently our strategy for AI Features is specific to .com and Ultimate users.
Logging interested customers
For GitLab team members, please register interested customers in this spreadsheet. This will help us conduct a structured prioritization between various customers.
Why is it hard to bring AI powered features to Self managed?
- AI Features can require specialized and expensive hardware like GPUs that are increasingly becoming hard to secure.
- GitLab's .com first approach to AI features is partially due to this requirement. GitLab has secured GPUs to power features for customers. We can spread the cost of running these GPUs across customers since they need to be leveraged 24/7/365.
- AI features are powered by AI models that need to run special inference servers that we don't expect customers have the expertise to setup and keep running.
- AI Models change continuously. We would need a way to push models to self managed customer instances. We currently do not have the capability to do this. It also increases the burden on customers to update these models.
- Time-to-Market: Innovating on SaaS is quicker given the ability to iterate and quickly update code (vs distributing patches to SM customers).
Internal guidance on SM features: https://gitlab.com/internal-handbook/internal-handbook.gitlab.io/-/merge_requests/2704/diffs
We are also exploring how we could leverage something like &308 to help bring AI features to self managed customers. Code suggestions (beta) will be available for self-managed customers beginning with 16.1. An internet connection will be required.
Edited by Roger Woo