Choose a skill for coding guidance, an MCP server for judgment tools, or a browser or model-routing integration.
These four projects connect Jev in different ways. Their documentation describes the capabilities below; installation and live behavior have not been tested here.
Use a documented agent skill to help a coding agent build System One workflows.
The skill can guide the coding agent toward typed Jev decisions and the official API surface.
The coding agent, local project and TypeSafe API still own credentials, code changes and side effects.
README reviewed. The skill helps design workflows; it does not install the hosted Jev model. Installation has not been tested here.
Expose a Jev judgment capability through an MCP server that an agent client can call.
The README describes claim checks, screening, classification, ranking and patch review tools returning structured judgments.
The MCP client and downstream application own tool permissions, credentials, validation and side effects.
README reviewed: Node.js 20+ and a TypeSafe API key are required. Not installed or tested here.
Put Jev inside an observe → choose → act → verify browser loop.
Jev chooses among serialized candidate actions or a stop / escalation result.
Browser Use owns navigation, waits, permissions, clicks, form input and result verification.
README reviewed: uv, Chrome, a TypeSafe key and a separate text-model key are required. No local performance test.
Use a bounded decision to select a model or reasoning tier before the coding agent continues.
Jev selects a model capability tier and reasoning effort from the router’s configured options.
The coding-agent runtime owns dispatch, fallback, permissions and the actual code operation.
README reviewed: this project provides a local routing service for Codex. It has not been installed or benchmarked here.