Configure and call AI models through the Olares Router gateway via olares-cli.
npx skills add https://github.com/beclab/Olares --skill olares-router
Please help me install this Agent Skill. Skill: olares-router Source: https://github.com/beclab/Olares/tree/main/cli/skills/olares-router Command: npx skills add https://github.com/beclab/Olares --skill olares-router
Managing AI models on a self-hosted Olares system requires coordinating cloud vendor credentials, local model applications, API keys, quotas, and routing defaults across two separate planes (the Router gateway and each model's own console). This Skill gives an AI agent the operational knowledge to perform all of that through the olares-cli router command family without guesswork.
Ask the agent to list the models currently configured on this Olares and show which model answers the default chat category.
High-intent search queries and answers about installing and using this skill.
Use olares-cli router provider create with the vendor type, base URL, and credentials passed via --credentials-json from a file or stdin, then import models with router model import. Add --validate to probe the upstream immediately so a wrong key is caught at setup.
Install a pinned model application with olares-cli market install, or clone an engine base (llama.cpp, vLLM, Ollama, SGLang) with market clone. Router automatically creates the provider row; track weight download with router model progress.
A running container and a working model are separate states: weights may still be downloading or loading after the app reports running. Check router model status and router model progress to see the model's own phase before changing any configuration.
No. Since Router v2.2.1, calls without an Authorization header are attributed to the active Olares profile identity. Keys are only needed for model allowlists, separate budgets, or callers outside the Olares platform.
Use router quota set scoped to a key, user, model, or caller application, with --budget for total dollar spend, --rpm/--tpm for rate limits, or --concurrent for in-flight calls. A ceiling of 0 blocks all calls in that scope without deleting anything.
It requires Olares 1.12.7+, an admin profile for most management verbs, and cannot manage application lifecycle operations like stop, resume, or GPU binding, which belong to olares-market and olares-settings. Router also has no idempotency keys, so timed-out writes must not be blindly retried.
beclabRead and mutate Olares Settings areas with role-based access checks.
DevMHOneExpose OpenAI-compatible inference endpoints for chat, embeddings, images, audio, and more.
scitixDiagnose Kubernetes StatefulSet rollout and scaling failures via pod ordinals, PVC bindings, and update strategies.
Ilenburg1993Manage the background Audit Agent, Inference Gateway, and ollama host supervisor for LLM operations.
