alt Hacker NewsDoes anyone do this without the vendor lock-in? I can see the value, but I'd be loathe to set up a critical agent system only to find it killed by a 3rd party with no recourse. Is there a self-hosted option?
Replies
I run a Matrix server with federation turned off. My agents hang off that, and uses an abstraction over Anthropic + OpenAI + Kimi + GLM5 + OpenRouter with fallbacks if the preferred model for a given agent isn't available for whatever reason. Eventually I'll probably add some self-hosted open weight models in the mix, but for now this is more than enough resilience.
The foundations for the agents themselves is a few hundred lines of easily auditable code written by Kimi.
Have a framework how the AI is allowed to work and keep that as a git repo.
That way you can tell AI 1 to write into it and AI 2 can see where every work you do is currently standing and continue from there.
Not sure what grok bit exactly offers, but OpenClaw or Hermes Agent seem similar.
I’m currently running Paseo for agent orchestration for code, beads managing tasks, and OpenWebUI for stuff that’s more on the business side (with RAG or MCP to our drive, CRM, etc). It’s mostly human triggered though it doesn’t have to be. Paperclip might be a more LLM-managed option.
All of it is self hosted, models are Ollama Cloud and OpenRouter. You can throw dirt cheap models at a lot of these tasks and they perform very well. It’d be no effort at all to replace inference providers.