Jev with AI agents
Let the chat model write prose. Let Jev decide which shell command, skill, route, or gate fires next.
Generative agents are brilliant at language and terrible at pretending to be a calibrated classifier. Jev answers fixed-shape questions with probabilities you can threshold in code.
These guides map verified integrations from the ecosystem: plugins, MCP servers, and agent skills that call POST /v1/systemone (or gateway equivalents) while your main model keeps the microphone.
Platform guides
Hermes Agent + Jev
Smart approvals, skill routing, and MCP without asking Hermes to fake a classifier.
Read guide →OpenClaw + Jev
Assistant OS from OpenClaw, decide layer from Jev. Routing, gates, and official decisionModel support.
Read guide →Claude Code, Desktop, and Cowork + Jev
Plugins, MCP connectors, and TypeSafe skills on Anthropic's agent stack.
Read guide →Codex and OpenCode + Jev
jev-judge on the CLI, evaluate setup, and routers that treat Jev as traffic control.
Read guide →Cursor, Grok, and agent skills + Jev
MCP in Cursor settings, TypeSafe skills in the agent, no imaginary Grok-only APIs.
Read guide →
Also works with
Patterns we did not give a full page yet, but the directory tracks them.
Native pi extensions and routers in the directory (y0usaf, TheoOliveira, mejiasd3v). typesafe-mcp ships evaluate setup pi.
Jev picks the browser operation; a small LLM types only when needed.
Community async supervision layer for Hermes (keeltrace/hermes-jev). Separate from anpicasso smart approvals.
Per-turn Codex model routing via Jev and Codex Router extensions.
FAQ
- Do I need a separate API key for every agent?
- Usually one TypeSafe key is enough. Export TYPESAFE_API_KEY in the shell or service that launches the agent (or reference it from MCP config with env interpolation where supported). Vercel AI Gateway users can use AI_GATEWAY_API_KEY for gateway-routed Jev instead; see the learn guide on Vercel AI Gateway Jev.
- MCP server or agent skill: which comes first?
- Skills teach the model when and how to design a Jev call. MCP servers expose tools like judge, evaluate, or jev_route that actually hit the API. Many teams install both: typesafe-ai/skills for patterns, plus a focused MCP package for the hot path.
- What happened to system1-mcp from Nous Research?
- There is no public NousResearch/system1-mcp repository in the indexes we checked. Hermes users typically wire Jev through Hermes plugins (approvals, skill routers), generic Hermes MCP support, or community MCP servers such as blakestone-x/jev-mcp and itsmostafa/typesafe-mcp.
- Is there a Muse plugin that uses Jev?
- We did not find a documented OpenClaw or TypeSafe integration branded Muse with Jev-specific tools. If you discover one, submit it to the directory. Until then, treat Muse as unrelated to Jev unless the upstream README names TypeSafe or System One explicitly.
- Where do Choice, Score, and Noul show up in agents?
- Approvals and routing use Choice (pick APPROVE/DENY/ESCALATE or a model route). Risk and severity use Score. Safety checks and policy fit use Noul (0 to 1). Parallel questions in one request are the default System One pattern; see /learn/system-one.
- Can Jev replace my main chat model?
- No. Jev does not draft user-facing paragraphs or long reasoning chains. It gates, ranks, classifies, and scores. Keep Claude, GPT, Grok, or Hermes' primary model for language; add Jev where you would otherwise regex-parse YES/NO from a completion.