Hermes Hindsight Guide: memory tool for Hermes Agent

Chinese guide to a local Hermes memory setup with Ollama and Hindsight that reduces memory clutter

What Hermes Hindsight Guide does

This guide targets a specific problem: with default settings, Hermes auto-retain writes irrelevant data into a Hindsight bank, auto-recall injects it into later conversations, and the agent drifts off topic. The author's answer is an external memory stack under your control. Ollama runs a local model, qwen2.5:7b-instruct-q4_K_M, for extraction, and the Hindsight API server stores a knowledge graph in PostgreSQL with pgvector, reached by Hermes over HTTP on localhost:9177. The README names Hermes v0.13.0 and Hindsight v0.6.1.

The tuned settings include retaining every five turns, a low recall budget, separate banks such as hermes-cli and hermes-telegram, raising the MEMORY.md limit from 2,200 to 5,000 characters, a bank mission that sets the topic boundary, and cleaning sessions after 14 days. The README reports MEMORY.md shrinking from 3,799 to 2,123 characters. It links SETUP.md, config-reference.md and architecture.md, and ships a GTK tray indicator script. Suggested hardware is 16 GB RAM minimum and 32 GB RAM with 16 GB VRAM recommended.

Key features

Local stack of Ollama, Hindsight API and PostgreSQL with pgvector
Tuned retain, recall and bank settings to cut irrelevant memories
Separate memory banks for Hermes CLI and Telegram
Step-by-step SETUP.md, config reference and architecture write-up
GTK system tray script that shows whether Hindsight is running
Hardware guidance for minimum and recommended setups

When to use it

Fixing a Hermes agent that mixes old, unrelated tasks into current conversations
Running Hermes long-term memory entirely on your own machine
Keeping CLI and Telegram memories from contaminating each other

Who it is for: Chinese-reading Hermes Agent users who use Hindsight memory and want it local and tidy.

How it fits with Hermes Agent

Written specifically for Hermes Agent's Hindsight memory integration, with config for ~/.hermes/hindsight/config.json and gateway restarts.

How to install Hermes Hindsight Guide

These commands are copied from the project's README. Check the repository for the latest steps before you run them.

ollama serve
ollama pull qwen2.5:7b-instruct-q4_K_M
pip install hindsight-api-slim

Requirements: Minimum 16 GB RAM and 8 GB VRAM or CPU-only; 32 GB RAM and 16 GB VRAM recommended; Ollama, Hindsight API and PostgreSQL

FAQ

What is Hermes Hindsight Guide?

Hermes Hindsight Guide is a Chinese guide for setting up a local Hermes Agent memory system with Ollama and Hindsight. It focuses on stopping memory pollution from default auto-retain and auto-recall.

How do I install Hermes Hindsight Guide?

The guide starts with ollama serve, pulling qwen2.5:7b-instruct-q4_K_M, and pip install hindsight-api-slim, then continues with a systemd service and Hermes config. The README tells you to read SETUP.md for the full steps.

Is Hermes Hindsight Guide free and open source?

The README says MIT, but the repository has no license file on GitHub, so default copyright applies. Check with the author before reusing the content.