Self-Hosted Honcho for Hermes Agent

What Self-Hosted Honcho for Hermes Agent does

Hermes Agent uses Honcho for its cross-session memory layer, and by default that runs on Plastic Labs' managed cloud with their Neuromancer models. This repository lets you host Honcho yourself with three configuration files on top of upstream Honcho, so no fork is required. It runs the full stack on your machine: the API, the Deriver, PostgreSQL with pgvector, and Redis.

LLM calls for memory work go to any OpenAI-compatible provider, with an optional backup provider, and the README names OpenRouter, Venice, Routstr, Together and Ollama as options. Your conversation data and user profile stay on your own hardware, and only inference requests leave it. The README compares three setups: managed cloud, self-hosted with an API, and self-hosted with a local model through Ollama or vLLM. It notes that general-purpose models may give slightly less precise deductive reasoning than Neuromancer, and that the main benefit here is data ownership. Setup is described as taking about 3 minutes.

Key features

• Runs Honcho's API, Deriver, PostgreSQL and Redis on your own machine
• Three config files on top of upstream Honcho, with no fork
• Primary and optional backup LLM provider through any OpenAI-compatible API
• Hermes Agent connects to the self-hosted API at localhost:8000
• Optional fully local setup with Ollama, vLLM or llama.cpp via LLM_VLLM_BASE_URL

When to use it

• Keeping Hermes Agent's cross-session user model off a third-party cloud
• Running Hermes memory on a VPS or home server with OpenRouter or Venice for inference
• Building a fully local memory stack with a model on your own network

Who it is for: Hermes Agent users who want its Honcho memory layer to live on their own server.

How it fits with Hermes Agent

It is built for Hermes Agent's Honcho-based memory layer and replaces the default Plastic Labs cloud endpoint with a self-hosted one.

Requirements: Ubuntu 22.04+ Linux server (tested on 22.04 with 6GB RAM) and an OpenAI-compatible LLM provider such as OpenRouter or Venice.

FAQ