Repository navigation
jev routing
mastalink edited this page Oct 9, 2026
·
1 revision
Use Jev — a deterministic, bounded System One decision provider — to make fast, structured routing decisions, recommend optimal model providers for tasks, and classify incoming work without authorizing side effects.
Jev provides programmable common sense as a core architectural primitive:
- Annotation-Only: Jev only classifies, ranks, and recommends. It never executes shell commands or writes database state directly.
- Structured Decisions: Turns natural language task descriptions into typed decisions (e.g., capability requirements, risk categories, optimal model selections).
- Zero Hallucination Routing: Evaluates objective criteria (latency, token costs, coding capability) to pick the right worker agent.
- Navigate to Settings → LLM Connections / Routing.
- Enter a task scenario:
- Example prompt: "Refactor payment gateway webhook retry logic with idempotency guarantees."
- The routing advisor analyzes the requirements and outputs:
-
Recommended Model:
claude-3-7-sonnet(high architectural reasoning) orcodex-davinci(code generation). -
Confidence Score: e.g.,
0.94. - Rationale: Explains why this model was chosen based on prompt complexity and coding requirements.
-
Recommended Model:
Run Jev decision queries directly from your shell:
# Query Jev model routing recommendation
mco jev route-model --task "Write comprehensive unit tests for local SQLite WAL concurrency"
# Inspect Jev operations health
curl http://127.0.0.1:18789/api/jev/health `
-H "Authorization: Bearer $env:MCO_AGENT_TOKEN"Jev outputs a structured JSON evaluation:
{
"recommendation": "codex",
"confidence": 0.95,
"reasoning": "Task involves local filesystem and unit test generation. High fit for codex worker.",
"fallback_role": "reviewer",
"risk_tier": "low"
}- Cause: Local Jev socket or model connection timed out.
-
Fix: BitCadence falls back gracefully to deterministic rule-based matching if Jev is unreachable. Verify connection in
mco doctor.
- Cause: Task prompt was ambiguous or lacked specific implementation details.
- Fix: Provide clearer instructions in the job description naming specific languages or frameworks.