ModelMarshal is a local-first AI control plane for JetBrains IDEs. It shows the model, estimated cost, data destination, selected context, and requested permissions before an AI request runs.
After you approve a request, ModelMarshal records a reproducible local receipt containing its configuration, approvals, results, estimated cost, and outcome.
Control AI inside your IDE
- Approval-first execution: Review the context, destination, permissions, and estimated cost before anything is transmitted.
- Local-first operation: Connect to a local Ollama-compatible endpoint and keep approved requests on your device.
- Provider choice: Use a supported OpenAI-compatible endpoint with your own provider account and credentials.
- Cost visibility: Review a planning estimate before using an external model.
- Reproducible receipts: Keep local records of prompts, approvals, responses, tests, costs, and outcomes.
- Local policies: Control data transmission and require additional confirmation for sensitive operations.
Designed for human review
ModelMarshal version 0.1.0 does not automatically execute shell commands, modify project files, change dependencies, alter databases, or deploy software. AI-generated content remains a suggestion until you review and use it.