VIDEO
HIGHLIGHT
Log InSign up
Self-Hosting Honcho: Free Local Memory for My Hermes Agent
SummaryTranscriptChat

Self-Hosting Honcho: Free Local Memory for My Hermes Agent

Introduction to Honcho and Self-Hosting

Overview of the Video

  • The speaker introduces their experience with Honcho, an external memory provider for the Hermes agent, highlighting its ability to remember personal details.
  • They express a need to transition from cloud hosting due to impending credit depletion and plan to self-host Honcho on their machine.
  • The video aims to cover what Honcho is, reasons for self-hosting, its functionality, and a live setup demonstration.

Understanding Honcho

What is Honcho?

  • Honcho is described as a memory infrastructure for AI agents developed by Plastic Labs, focusing on continual learning and better context management.
  • It operates under the concept of "theory of mind," allowing AI agents to retain facts about users and model their working styles.
  • Users can access Honcho through managed services or by self-hosting via FastAPI server; the latter will be demonstrated in this video.

How Does Honcho Work?

Mechanism of Operation

  • After each conversation, Honcho runs a dialectic process that analyzes chat content to derive structured notes about user preferences and goals without interrupting ongoing chats.
  • At query time, it performs semantic searches while incorporating a compact model of the user’s identity into responses. Benchmarks show high performance scores in memory quality metrics like long mem S (90.4%) and LocoMo (89.9%).

Integration with Hermes Agent

Role of External Memory Providers

  • The Hermes agent has built-in persistent memory but also supports external providers like Honcho for deeper memory capabilities beyond basic notes stored in two files: memory.md (agent's notes) and user markdown (user profile).
  • External providers enhance functionalities such as knowledge graphs, semantic search, automatic fact extraction, and cross-session modeling while running alongside built-in memory systems without replacement.

Setting Up Local Environment

Prerequisites for Setup

  • Essential prerequisites include having the Hermes agent operational along with Git, Docker, Docker Compose installed on your machine; an LLM for reasoning may also be required depending on whether you choose local or cloud options for processing power.

Live Setup Process

  • The speaker demonstrates cloning the Honcho repository and setting up containers using Docker Compose which includes Redis and database services among others necessary for operation.

Configuration Details

Understanding Configuration Files

  • A detailed walkthrough of the docker-compose.yml file reveals four main services: API service running on port 8000; driver service dependent on cache; database service; Prometheus monitoring option available if uncommented in configuration files.

Environment Variables Setup

  • The .env file contains crucial settings including log levels, DB connection URIs for Postgres with PG vector support; authentication settings are advised against enabling unless in production environments due to security concerns.

Troubleshooting Connection Issues

Debugging Connection Problems

  • Initial attempts at saving conclusions fail due to misconfigured base URLs when connecting between different devices (Mac vs Raspberry Pi), necessitating IP address adjustments instead of localhost references.
  • Further issues arise from workspace mismatches leading to errors when trying to save conclusions until proper embedding settings are enabled.

Finalizing Setup & Testing Functionality

Successful Integration Confirmation

  • After resolving embedding dimension requirements by selecting compatible models that support necessary dimensions (1536), successful conclusion saving indicates that local setup functions correctly.
  • Speaker plans to share their repository containing configurations used during setup so others can replicate without facing similar challenges encountered throughout this process.

This structured summary captures key insights from the transcript while providing timestamps linked directly back to specific moments in the video for easy reference.

Video description

My Hermes agent has actually known who I am for three months and I just realized the cloud credits keeping that memory alive are about to run out, which left me with two options: keep paying for hosted memory, or self-host the whole thing myself. I went with self-hosting, so in this video I move my AI agent's entire memory stack off the cloud and onto my own machine, running it fully locally with Ollama for $0. Fair warning: there are almost no guides for this, and it cost me close to two to three hours of errors to get working, so I hand you the exact config. This is the full, honest walkthrough: what Honcho is, why I wanted it off the cloud, how it works under the hood, how it fits alongside my Hermes agent's built-in memory, and a live run at the end to prove the whole stack runs on my Mac. And I'm not hiding the messy part. Getting the reasoning and embeddings running on local models is where everything broke, and I show you exactly what I changed so you can just save some time. One thing to be straight about: starting local means the memory begins from a blank slate (I'm not migrating my cloud history), but it's free, it's fully mine, and it rebuilds as I use it. What you'll get out of it: • Whether self-hosting agent memory is worth it for you (cost, privacy, ownership) • Standing up Honcho locally with Docker + FastAPI • Running reasoning + embeddings on local models via Ollama, so nothing leaves your machine • Wiring Honcho in as your agent's external memory provider, plus the exact embedding setting that makes conclusions actually save If you're paying for a hosted memory provider and watching the credits tick down, this is the setup I'd try first. I've pushed my full config (docker-compose plus every environment variable) so you can clone it and skip the trial-and-error. 0:00 Intro: I've used Honcho for 3 months 0:55 Why I want to self-host (cloud credits running out) 1:33 What is Honcho? 2:40 How it works: the deriver + dialectic reasoning 3:18 Benchmarks 3:38 How Honcho fits with my Hermes agent (built-in vs external memory) 9:11 Prerequisites 10:04 Clone + Docker 11:02 docker-compose & .env walkthrough 13:48 Running the LLM locally with Ollama 16:18 Rebuild & containers healthy 17:00 Health check + smoke test 17:45 Connecting Honcho to Hermes 19:43 Debugging live: connection refused, 404, 500 23:36 The root cause: embeddings + 1,536 dimensions 26:42 It works: fully local 28:58 Outro + grab the repo Links • My setup repo (config + env vars): https://github.com/nidhi-singh02/honcho-local-setup • Honcho (GitHub): https://github.com/plastic-labs/honcho • Ollama: https://ollama.com If you're paying for a hosted memory provider, run this test before your next invoice. And if it saved you time, subscribe: it genuinely helps a small creator keep putting work like this out. #Honcho #AIAgents #AgentMemory #SelfHosted #Ollama