Inference
Zenith route-lease
Agenticaria owns agents. Zenith owns model routing. When you connect them, Agenticaria requests a lease before chat — then selects the matching artifact and runs tools locally.
Agenticaria
Agent plane
- Blueprints & compile
- Prompt artifacts
- Eval & promote
- Tools & receipts
- Run API & widget
Zenith
Inference plane
- Provider pools
- Smart routing
- Route lease
- Fleet policy
- Residency controls
Agenticaria asks Zenith for a lease, then runs the matching artifact — Zenith never stores prompts or executes tools.
Boundary
Agenticaria does
- Compile blueprints into model-specific artifacts
- Eval and promote versions
- Execute tools and keep receipts
- Serve Run API and widget embeds
Zenith does
- Pool providers and route traffic
- Issue route leases before chat
- Enforce fleet policy and residency
- Never store prompts or run tools
Enable lease mode
Point inference at a Zenith OpenAI-compatible base and turn lease on. On retry_lease, the runtime refreshes and retries.
# inference via Zenith gateway
AGENTICARIA_INFERENCE_MODE=zenith_gateway
AGENTICARIA_INFERENCE_BASE_URL=https://zenith.example/v1
AGENTICARIA_INFERENCE_API_KEY=zk_…
AGENTICARIA_ZENITH_ROUTE_LEASE=trueStandalone still works Skip Zenith entirely with
AGENTICARIA_INFERENCE_MODE=standalone and any OpenAI-compatible upstream.