---

name: routers-services description: "Use ClearML HTTP routers, local FastAPI proxies, endpoint telemetry, and long-running service patterns safely." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

ClearML Routers and Services

Use this sub-skill when a user wants to proxy a local HTTP model or web service through ClearML, diagnose HttpRouter import/setup failures, design request/response callbacks, register endpoint telemetry, or package a service/monitor/autoscaler pattern.

Start Here

• For API names, signatures, optional dependency notes, callback contracts, and endpoint-return shapes, read references/api-reference.md.
• For local proxy plans, telemetry choices, deploy/wait/list/remove sequences, static routes, and service-task patterns, read references/workflows.md.
• For clearml[router] failures, FastAPI/uvicorn/httpx import errors, port conflicts, callback bugs, endpoint wait timeouts, and service safety, read references/troubleshooting.md.
• To check whether the optional router dependencies are importable without starting a server, run scripts/router_extra_check.py.
• To generate a safe code plan for a local route and callbacks without binding a port or contacting a ClearML server, run scripts/router_plan.py.

Route Selection

• Use this sub-skill for Task.get_http_router(), HttpRouter.set_local_proxy_parameters(), create_local_route(), deploy(), wait_for_external_endpoint(), list_external_endpoints(), local FastAPI proxy behavior, request/response/error callbacks, and endpoint telemetry.
• Use ../experiment-tracking/SKILL.md when the main question is Task creation, logging, parameters, artifacts, models, or credentials outside router/service serving.
• Use ../remote-execution-cli/SKILL.md when the main question is launching a service on a ClearML Agent queue, CLI task creation, queue selection, or Task.execute_remotely() behavior.
• Use ../automation-pipelines/SKILL.md when the main question is schedulers, automation controllers, pipelines, optimizers, or broader orchestration.

Safety Defaults

• Treat router deployment as a live operation: create_local_route() and deploy() can start a uvicorn-backed proxy subprocess, bind a local port, and request an external endpoint from the ClearML server.
• For local dry runs, set endpoint_telemetry=False, choose an unused incoming_port, and generate a plan with scripts/router_plan.py before running live code.
• Do not put API keys, Slack tokens, cloud credentials, or server configuration in generated snippets; use ClearML configuration and secret handling outside the router code.