← Back to Skills Marketplace
Ollama Herd
by
Twin Geeks
· GitHub ↗
· v1.5.3
· MIT-0
245
Downloads
0
Stars
3
Active Installs
9
Versions
Install in OpenClaw
/install ollama-herd
Description
Ollama multimodal model router for Llama, Qwen, DeepSeek, Phi, and Mistral — plus mflux image generation, speech-to-text, and embeddings. Self-hosted Ollama...
Usage Guidance
This skill appears to be a legitimate Ollama fleet manager but ask yourself and verify the following before installing: 1) The SKILL.md directs 'pip install ollama-herd' and running daemon processes — review the PyPI package and the linked GitHub source (compare code, maintainer, recent activity, and release checksums) before installing. 2) Confirm the localhost API (port 11435) is intended to be unauthenticated or will be bound to loopback only; if it is exposed more widely it could be abused. 3) The SKILL.md metadata lists local config files (~/.fleet-manager/latency.db and logs) even though the registry summary did not — clarify whether the skill will read/write those files and whether they may contain sensitive data. 4) Run initial tests in an isolated VM/container if possible, and back up any existing Ollama or fleet config. 5) If you need higher assurance, request a packaged install spec (pinned version, checksum) or a copy of the package contents so you or a reviewer can inspect what the installed code does (network calls, file I/O, subprocesses). If the maintainer can confirm the security model for the HTTP API and publish a verifiable release, my confidence would increase.
Capability Analysis
Type: OpenClaw Skill
Name: ollama-herd
Version: 1.5.3
The ollama-herd skill bundle is a legitimate management tool for a distributed Ollama AI fleet. It provides instructions and commands for monitoring node health, managing models, and querying local performance metrics via a local API (localhost:11435) and a dedicated SQLite database (~/.fleet-manager/latency.db). The bundle includes clear guardrails for the agent and lacks any indicators of data exfiltration, malicious execution, or unauthorized access.
Capability Assessment
Purpose & Capability
The name/description (an Ollama multimodal router/fleet manager) aligns with the SKILL.md content: it documents endpoints for fleet status, model pulls, health checks, etc. Requiring curl/wget and optionally python/pip/sqlite3 is reasonable for a CLI/HTTP-based local manager. However, the registry summary earlier listed no required config paths while the SKILL.md metadata declares configPaths (~/.fleet-manager/latency.db and ~/.fleet-manager/logs/herd.jsonl) — that discrepancy is unexplained and suggests the skill expects access to local files not declared in the registry.
Instruction Scope
SKILL.md instructs the agent and user to pip install a PyPI package (ollama-herd), run herd and herd-node, and call many localhost:11435 endpoints (GET/POST) that can change router settings and manage models. Those actions are coherent with a fleet manager but are powerful: they install software, start services, and perform model pulls/deletes. The instructions reference local config/log paths in metadata although they never explicitly show reading them; this raises scope questions (will the agent read or modify those files?). Also the instructions assume an unauthenticated or local HTTP API at port 11435 — the security model for that API is not specified.
Install Mechanism
There is no formal install spec in the registry (instruction-only), but the SKILL.md instructs users/agents to run pip install ollama-herd from PyPI. Installing a third‑party PyPI package and running its daemons is a moderate-risk action — it executes upstream code on the host. Because the registry provides no pinned release, checksum, or local package bundle, you should verify the PyPI package and upstream source before installing.
Credentials
The skill requests no environment variables or credentials, which is proportionate. It does, however, imply access to local ports and files (the metadata configPaths). The lack of declared required config paths in the registry vs. their presence in SKILL.md metadata is inconsistent and could hide file access expectations. No credentials are requested, but the API endpoints called can affect system state (pull/delete models), so local process permissions are effectively required.
Persistence & Privilege
The skill is not force-enabled (always: false) and permits normal autonomous invocation. It does not request to modify other skills or system-wide agent settings in the registry. The main persistence/privilege concern is operational: installing the package and running 'herd-node' will create long-running services and open local ports — appropriate for a fleet manager but something the user must opt into.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install ollama-herd - After installation, invoke the skill by name or use
/ollama-herd - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.5.3
Added /api/pull endpoint docs and examples
v1.5.2
Cross-platform support: macOS, Linux, and Windows. Updated OS metadata, descriptions, and hardware recommendations.
v1.5.1
**This release refines references and documentation for clarity and Ollama-specific focus.**
- Updated documentation to refer explicitly and consistently to "Ollama" across all sections, endpoints, and feature descriptions.
- Added multilingual tagline and brief descriptions in Chinese and Spanish for broader accessibility.
- Clarified example curl commands and usage descriptions to explicitly mention "Ollama" models and features.
- Improved installation and dashboard usage instructions to highlight Ollama usage and compatibility.
- No functional or code changes; documentation only.
v1.5.0
- Updated version to 1.5.0.
- Modified the metadata field format for improved parsing and compatibility.
- No changes to endpoints, usage, or documentation content.
v1.4.0
- Simplified description: now emphasizes Apple Silicon and routing for Mac Studio, Mac Mini, MacBook Pro.
- Minor description tweaks for clarity and updated model/router capabilities.
- No command, feature, or endpoint changes; all usage, dashboard, and curl/API instructions remain the same.
- Technical detail level and core documentation are unchanged.
v1.3.0
- Updated description to highlight Llama, Qwen, DeepSeek, Phi, and Mistral model support, plus multimodal routing.
- Clarified Apple Silicon focus and inference routing features in the description.
- No functional or API changes documented.
- Version remains at 1.1.0.
v1.2.0
Multimodal support and improvements for fleet routing and management:
- Adds support for 4 model types: LLM inference (Ollama), image generation (mflux), speech-to-text (Qwen3-ASR), and embeddings.
- Updates the description and router details to clarify multimodal capabilities.
- Maintains API endpoints, dashboard features, and resilience mechanisms.
- Metadata format updated for OS requirements.
- The rest of the guide and usage examples remain consistent with previous versions.
v1.1.1
- Added "optionalBins" (python3, sqlite3, pip) and "configPaths" (~/.fleet-manager/latency.db, ~/.fleet-manager/logs/herd.jsonl) entries to metadata for improved environment detection.
- No user-facing functional or documentation changes.
v1.1.0
- Adds comprehensive device fleet management features, including node status checks, queue monitoring, and model insights.
- New dashboard with eight tabs for live fleet status, analytics, health checks, recommendations, and settings.
- Introduces API endpoints for fleet health analysis, usage stats, model management (pull/delete), and request tracing.
- Includes resilience features: auto-retry, model fallback, VRAM/cold-load awareness, and zombie task cleanup.
- Enhanced recommendations and per-app analytics for optimal model distribution and usage tracking.
- Expanded documentation with clear curl commands and troubleshooting workflows.
Metadata
Frequently Asked Questions
What is Ollama Herd?
Ollama multimodal model router for Llama, Qwen, DeepSeek, Phi, and Mistral — plus mflux image generation, speech-to-text, and embeddings. Self-hosted Ollama... It is an AI Agent Skill for Claude Code / OpenClaw, with 245 downloads so far.
How do I install Ollama Herd?
Run "/install ollama-herd" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Ollama Herd free?
Yes, Ollama Herd is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Ollama Herd support?
Ollama Herd is cross-platform and runs anywhere OpenClaw / Claude Code is available (darwin, linux, windows).
Who created Ollama Herd?
It is built and maintained by Twin Geeks (@twinsgeeks); the current version is v1.5.3.
More Skills