What is Hermes OKF?

 

Hermes OKF gives your AI agent a persistent, structured, version-controlled memory — no database, no lock-in, just markdown + YAML on your filesystem. Every decision, observation, and project context lives in a human-readable knowledge graph that your agent can read, write, and traverse programmatically.

pip install hermes-okf

Some of the CLI commands are below 

# Initialise a new OKF bundle

hermes-okf init ./knowledge

# Validate conformance

hermes-okf validate --path ./knowledge

# List concepts

hermes-okf list --path ./knowledge

# Show a concept

hermes-okf show --path ./knowledge projects/my_project

# Search

hermes-okf search --path ./knowledge "ffmpeg GPU"

# View log

hermes-okf log --path ./knowledge

# Append to log

hermes-okf log-append --path ./knowledge "New decision made" --category Decision

# Graph inspection

hermes-okf graph-edges --path ./knowledge

hermes-okf graph-neighbors --path ./knowledge projects/my_project

# Save snapshot

hermes-okf snapshot --path ./knowledge --note "Before deploy"

# Build LLM context

hermes-okf context --path ./knowledge "What should I prioritize?"

# List sessions, plans, tools

hermes-okf sessions --path ./knowledge

hermes-okf plans --path ./knowledge

hermes-okf tools --path ./knowledge

Feature What You Get

🧠 Agent Memory Persistent decisions, observations, and tool-call history across sessions

🔗 Knowledge Graph Implicit graph from markdown links — no RDF, no Cypher

📁 Filesystem-First Plain .md + YAML. cat it, grep it, Git it.

⚡ Zero-DB Core Single dependency: pyyaml. Optional RAG via LangChain/ChromaDB.

🔌 Hermes Plugin HermesOKFMemoryProvider — native MemoryProvider ABC, discovered via hermes-okf install-plugin

🎁 Hermes-Ready Drop-in decorators: @memorize_decision, @memorize_tool

🔄 Resume Stop and restart — the agent restores from its OKF bundle

📦 Portable Clone a bundle to another machine — the agent resumes instantly.

OKF (Open Knowledge Format) is a vendor-neutral, open specification published by Google Cloud on June 12, 2026. It formalizes the "LLM wiki" pattern into a portable standard: every concept is a .md file with YAML frontmatter, and markdown links create a knowledge graph.

"OKF is a vendor-neutral, agent- and human-friendly standard for representing the metadata, context, and curated knowledge that modern AI systems need." — Sam McVeety & Amir Hormati, Google Cloud

Why hermes-okf chose OKF:

OKF Principle What it means for agents

Minimally opinionated Only one required field: type. Everything else is up to the producer.

Producer/consumer independence A human can write a bundle; an AI agent can read it. No lock-in.

Format, not platform No proprietary runtime, no SDK, no cloud required. Just markdown files.

Human-readable cat any file and understand it. Git diffs work out of the box.