self-hostedai

Llama Rider

· Vitor Pontual · 1 min read

Autonomous AI agent runtime inspired by Open Claw, rebuilt from scratch to solve the problem of 100% Ollama compatibility—no OpenAI API shims, no cloud dependencies. Powered by a custom 80B parameter model routed through the Ollama Fleet proxy to a DGX Spark, accessible entirely through a Telegram bot interface.

  • 16 built-in tools — shell execution, filesystem ops, Docker management, web fetching, Git, SearXNG search, persistent memory, and more
  • ReAct loop — async generator-based think-act cycle that chains tools autonomously to solve multi-step tasks
  • Skill system — installable SKILL.md files (OpenClaw format) for extending the agent with new capabilities
  • Session persistence — conversation context survives restarts with automatic compaction when context window fills
  • Custom 80B modelqwen3-coder-next:tools with hand-grafted tool-calling templates for native Ollama tool use
  • Zero cloud dependencies — runs entirely on local hardware, all inference through your own Ollama fleet