Connect Ollama, LM Studio, or a compatible model endpoint to Flame IDE. Use local AI with project context, configurable AI surfaces, and your existing tools.
Running inference locally is only part of a useful coding workflow. You still need the right files, current changes, terminal output, and a place to inspect the result. Flame connects local model providers to the workspace around your repositories. Choose where local AI fits, keep provider selection explicit, and continue using Git, tests, terminals, and project tools in the same IDE.
Flame supports Ollama, LM Studio, and OpenAI-compatible provider profiles. Local endpoint discovery can probe the usual Ollama and LM Studio addresses, while explicit profiles let you configure the server you intend to use.
Configure AI surface defaults rather than assuming one model is best everywhere. Try a local model for chat, completion, or a focused AI action, and verify that the chosen model supports that surface's input and response format.
Use project files and selected code as context, inspect the proposed result, and run the same tests you would use for any change. Local inference does not require a separate copy of the repository or a separate review process.
Install and start Ollama or LM Studio using its official instructions. Download a model suited to your hardware and task, then confirm that its local API server is running. Model downloads and installation can require internet access.
Open Settings and find AI Providers. Use local endpoint discovery or add the appropriate provider profile. Confirm the base URL, enable the profile, and load the models exposed by that server. The endpoint must be reachable from the computer running Flame's backend.
Choose the local provider and model in the relevant AI control or surface default. Start with a small project question or focused code action. Check the actual provider selection before sending code context.
Review the result against the source and run relevant validation. If the model is too slow or cannot handle the context, reduce the task size, review its loaded context setting, or choose a model that fits the available memory.