ClawHub

AI Agent Skill

Give your AI agent instant knowledge of your fleet.

Ollama Herd ships with a ClawHub skill that gives AI agents instant knowledge of your fleet. Install it once and any OpenClaw-compatible agent can check node health, route requests, pull models, run diagnostics, and manage your fleet, all through natural language.

What the Skill Does

The Ollama Herd skill teaches your AI agent:

  • Fleet operations, check node status, queue depths, loaded models, health scores
  • Request routing, send LLM, embedding, image generation, and speech-to-text requests through the fleet
  • Model management, pull models to specific nodes, list available models, check hot vs. cold state
  • Diagnostics, query the trace database, read JSONL logs, identify slow requests and bottlenecks
  • Health monitoring, 30+ automated health checks covering offline nodes, memory pressure, thermal throttling, model thrashing, error rates, and more
  • Dashboard access, point users to the real-time web dashboard for visual monitoring

Install

clawhub install ollama-herd

Or search for it on ClawHub:

clawhub search "ollama herd"

What Your Agent Can Do After Install

Check fleet health:

"How's my Ollama fleet doing?"

The agent hits /fleet/status and /dashboard/api/health, then summarizes node status, loaded models, queue depths, and any health warnings.

Route a request:

"Ask llama3.3:70b to explain quicksort"

The agent sends the request through the fleet router. Herd picks the best node automatically.

Pull a model:

"Pull codestral onto the Mac Studio"

The agent calls /api/pull with the target node. Streams progress in real time.

Diagnose issues:

"Why were my responses slow yesterday?"

The agent queries the trace database for high-latency entries, checks for thermal throttling or memory pressure events, and reports what it finds.

Get recommendations:

"What models should I run on each machine?"

The agent hits /dashboard/api/recommendations for AI-powered model mix suggestions based on your hardware and usage patterns.

Supported Platforms

The skill works with Ollama Herd on macOS, Linux, and Windows. Platform-specific features (image generation, speech-to-text, meeting detection) are noted in the skill and the agent adapts its suggestions accordingly.

API Endpoints the Skill Uses

Endpoint Purpose
GET /fleet/status Full fleet state, nodes, models, queues
GET /fleet/queue Lightweight queue depths for backoff
POST /api/chat Ollama-format chat completions
POST /v1/chat/completions OpenAI-format chat completions
POST /api/pull Pull models onto fleet nodes
GET /api/tags List all available models
GET /api/ps List models currently loaded in memory
POST /api/embed Generate embeddings
GET /dashboard/api/health 30+ automated health checks
GET /dashboard/api/traces Recent request traces with scoring details
GET /dashboard/api/recommendations Model mix recommendations per node
GET /dashboard/api/usage Per-node, per-model usage statistics
GET /dashboard/api/settings Current fleet configuration

Source