Skip to content
v0.1.2MIT

agent-tools-runtime plugin for Ollama's REST API: list/pull models, single-shot generate/chat completions, and async jobs (start_generate/start_chat + job_status) for calls that take a while, against a local server or Ollama Cloud.

What this package declares

The file a client reads when it loads this plugin, exactly as this revision carries it.

plugin.json
{
  "$schema": "https://agent-plugins.org/schemas/1.0.0/plugin.schema.json",
  "name": "ollama",
  "version": "0.1.2",
  "description": "agent-tools-runtime plugin for Ollama's REST API: list/pull models, single-shot generate/chat completions, and async jobs (start_generate/start_chat + job_status) for calls that take a while, against a local server or Ollama Cloud.",
  "license": "MIT",
  "repository": "https://github.com/MauricioPerera/agent-tools-runtime",
  "keywords": ["agent-tools-runtime", "rest-api", "ollama", "llm"],
  "extensions": {
    "dev.agent-tools-runtime": {
      "prefix": "ollama",
      "adapter": "./adapter.mjs",
      "adapterExport": "OllamaAdapter",
      "readonlyTools": ["list_models", "list_running_models", "job_status", "list_jobs"],
      "skills": ["./skills/vision_chat.mjs", "./skills/ocr.mjs", "./skills/grounding.mjs"],
      "discoverHint": "Por default apunta a un server de Ollama local (OLLAMA_URL, default http://localhost:11434, sin auth). Si se configura OLLAMA_API_KEY sin OLLAMA_URL, apunta a Ollama Cloud (https://ollama.com) en vez de requerir Ollama instalado -- ver https://docs.ollama.com/cloud. generate/chat siempre corren con stream:false y requieren confirm. Para modelos que puedan tardar (cloud grande, modelos que razonan mucho), usá start_generate/start_chat en vez de generate/chat: devuelven un jobId al toque y consultás el resultado con job_status -- no bloquean la llamada. El tracking de jobs vive en memoria del proceso runtime, se pierde si se reinicia. Las skills vision_chat/ocr/grounding envuelven chat() con un solo mensaje + imagen y un prompt fijo por tarea -- ninguna reemplaza a QwenLM/Qwen-MM-Plugins (esa API llama a Qwen vía DashScope), son el equivalente funcional armado sobre gemma4:cloud vía Ollama, verificado en vivo con imágenes reales antes de escribirse, no asumido. Las tres exigen confirm:true (mismo motivo que chat/generate: cómputo real) y un modelo con capability \"vision\" (ver list_models) -- sin esa capability Ollama ignora la imagen en silencio."
    }
  }
}

What else this package ships

These files come with the package and this site does not publish them. They are listed so you know what is there before you install it.

  • 4JavaScript files
  • LICENSE
View on GitHub

Client extensions

Data this package carries for particular clients. The directory lists the clients named and never reads what is addressed to them.

  • dev.agent-tools-runtime