hermes-okf 0.3.7

Universal OKF-based memory system for Hermes agent — structured, persistent, agent-readable knowledge storage.

Hermes OKF (Open Knowledge Format) — Universal Memory for AI Agents

The first open-source memory system built on Google's Open Knowledge Format (OKF) for the Hermes agent ecosystem. v0.3.7 syncs the OKF config concept with the actual Hermes model from config.yaml. HermesOKFMemoryProvider — a native Hermes Agent plugin with MemoryProvider ABC integration. pip install hermes-okf and Hermes auto-discovers it. hermes okf search|list|show|snapshot|restore CLI commands work out of the box.

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.

---

Why Hermes OKF?

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, auto-discovered after pip install
🎁 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.

---

Why OKF?

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.

References:

• Google Cloud Blog: Introducing the Open Knowledge Format
• OKF Specification (GitHub)
• Google Cloud Knowledge Catalog

---

Quick Start

As a Hermes Agent Plugin (Recommended)

pip install hermes-okf

Add to ~/.hermes/config.yaml:

plugins:
  enabled:
    - hermes-okf

memory:
  provider: hermes-okf
  bundle_path: ~/.hermes/okf_memory
  agent_id: hermes-alpha

Important: plugins.enabled must be a YAML list, not a string. If you use hermes config set plugins.enabled '["hermes-okf"]', it stores a JSON string which Hermes ignores. Edit ~/.hermes/config.yaml directly to ensure it's a proper list.

Run the setup wizard:

hermes memory setup

Hermes auto-discovers the plugin via pip entry points. No code changes needed.

As a Standalone Library

from hermes_okf import HermesOKFProvider

provider = HermesOKFProvider()
provider.on_session_start("session-1")
provider.on_memory_write("user", "User prefers dark mode")
provider.on_tool_call("search_web", {"query": "Python"}, "Found 5 results")
provider.on_decision("Use Claude", "Better reasoning", tags=["model"])
provider.on_session_end("session-1")

---

Hermes Plugin CLI

When installed as a Hermes plugin, these subcommands are available:

# Search your OKF memory
hermes okf search "dark mode"

# List stored concepts
hermes okf list --type Decision

# Save a snapshot
hermes okf snapshot --note "Before deployment"

# Restore from last snapshot
hermes okf restore

---

Agent Integration (Memory Mixin)

For most Hermes users, the plugin approach above is recommended. The decorators below are for advanced use cases or custom agent classes.

from hermes_okf.agent import HermesMemoryMixin

class MyAgent(HermesMemoryMixin):
    def __init__(self):
        super().__init__("./agent_knowledge", agent_id="my-agent-v1")
        self.start_session()

    @HermesMemoryMixin.memorize_decision
    def choose_model(self, task: str) -> str:
        if "code" in task.lower():
            return "anthropic/claude-3.5-sonnet"
        return "openai/gpt-4o"

    @HermesMemoryMixin.memorize_tool
    def scrape_data(self, url: str) -> dict:
        return {"url": url, "items": 42}

# Run it
agent = MyAgent()
agent.choose_model("Write a Python script")
agent.scrape_data("https://example.com")

# Recall relevant context
context = agent.with_context("python script", top_k=3)

---

Full Hermes Agent (State as OKF Bundle)

For deeper integration, use HermesAgent — the entire agent state lives in the OKF bundle. The agent can be stopped, restarted, and resumed from its bundle alone.

from hermes_okf import HermesAgent

# Create or resume an agent
agent = HermesAgent(
    bundle_path="./hermes_agent_brain",
    agent_id="hermes-alpha",
    model="anthropic/claude-3.5-sonnet",
)

# Register tools with JSON schemas
agent.register_tool("search_web", "Search the web", schema={"type": "object", ...})

# Create and execute plans
agent.create_plan("Research AI trends", ["Search", "Summarize", "Report"])
agent.complete_step(0, result="Found 5 major trends")

# Build LLM context automatically
context = agent.build_context("What should I do next?")

# Save snapshot — resume later from this exact state
agent.snapshot()

agent.end_session()

The agent bundle structure:

hermes_agent_brain/
├── config/agent.md        # Identity, model, system prompt
├── tools/*.md             # Tool definitions with schemas
├── sessions/*.md          # Session records
├── plans/*.md             # Active plans with checkable steps
├── plans/archive/*.md     # Completed plans
├── decisions/*.md         # Strategic decisions
├── snapshots/*.md         # Full state snapshots
└── index.md / log.md      # Bundle overview and activity log

Read the full integration guide: docs/HERMES_INTEGRATION.md

For the universal provider usage, see: docs/HERMES_USERS.md

---

Standalone CLI

# 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

---

RAG Integration (Optional)

pip install hermes-okf[rag]

Feed your OKF bundle into LangChain + ChromaDB for vector retrieval:

from langchain_community.document_loaders import DirectoryLoader, TextLoader
from langchain.text_splitter import MarkdownHeaderTextSplitter
from langchain_chroma import Chroma
from langchain_openai import OpenAIEmbeddings

from hermes_okf.bundle import OKFBundle

bundle = OKFBundle("./my_knowledge")
loader = DirectoryLoader(
    str(bundle.root),
    glob="**/*.md",
    loader_cls=TextLoader,
    loader_kwargs={"encoding": "utf-8"},
)
docs = loader.load()

splitter = MarkdownHeaderTextSplitter(
    headers_to_split_on=[("#", "Header 1"), ("##", "Header 2")]
)
splits = [chunk for doc in docs for chunk in splitter.split_text(doc.page_content)]

vectorstore = Chroma.from_documents(
    documents=splits,
    embedding=OpenAIEmbeddings(),
    persist_directory="./chroma_okf_db",
)

# Query
retriever = vectorstore.as_retriever(search_kwargs={"k": 5})
results = retriever.invoke("What GPU decisions did we make?")

See examples/rag_integration.py for a complete example.

---

Architecture

┌─────────────────────────────────────┐
│  CLI (hermes-okf)                   │
├─────────────────────────────────────┤
│  HermesOKFMemoryProvider            │  ← Hermes plugin (v0.3.1)
│  HermesOKFProvider                  │  ← Universal provider (v0.3.0)
│  HermesAgent / MemoryMixin          │  ← Agent integration
├─────────────────────────────────────┤
│  OKFBundle                          │  ← Core read/write API
│  ├── Concept (dataclass)            │
│  ├── GraphExtractor                 │
│  ├── SearchIndex                    │
│  └── OKFValidator                   │
├─────────────────────────────────────┤
│  Filesystem (markdown + YAML)       │  ← Persistent storage
└─────────────────────────────────────┘

Read the full architecture in docs/ARCHITECTURE.md.

---

OKF Format

This library implements the Google Open Knowledge Format (OKF) v0.1 spec:

• Every concept is a .md file with YAML frontmatter
• Frontmatter must include a type field
• Reserved files: index.md (directory), log.md (chronology)
• Markdown links create implicit directed edges
• Directory structure provides containment hierarchy
• Tags create cross-cutting clusters

"If you can cat a file, you can read OKF."

---

Project Structure

hermes-okf/
├── src/hermes_okf/          # Core library
│   ├── bundle.py             # OKFBundle read/write
│   ├── concept.py            # Concept dataclass
│   ├── graph.py              # Graph extraction & traversal
│   ├── search.py             # Full-text search & indexing
│   ├── validators.py         # OKF conformance checking
│   ├── memory.py             # Agent memory layer
│   ├── agent.py              # Drop-in decorators
│   ├── hermes.py             # HermesAgent full-state bundle
│   ├── hermes_integration.py # Universal HermesOKFProvider (v0.3.0)
│   ├── memory_plugin.py      # HermesOKFMemoryProvider plugin (v0.3.1)
│   └── cli.py                # CLI entry point
├── tests/                    # pytest suite
├── examples/                 # Usage examples
├── docs/                     # Architecture docs
└── .github/workflows/        # CI/CD

---

Development

git clone https://github.com/EliaszDev/hermes-okf.git
cd hermes-okf
pip install -e ".[dev]"
pytest

See CONTRIBUTING.md for full guidelines.

---

Roadmap

• Hermes plugin (HermesOKFMemoryProvider) — MemoryProvider ABC, auto-discovered, hermes memory setup integration
• Universal Hermes memory provider (HermesOKFProvider) — any Hermes agent can use it
• Two-memory model (hot buffer + cold OKF archive) with automatic flushing
• Hermes config system integration (~/.hermes/hermes-okf.yaml)
• Async I/O support for high-throughput agents
• Multi-agent bundle merging and conflict resolution
• Git-backed history with automatic diff summaries
• Web viewer for knowledge graph exploration
• Plugin system for custom concept types and validators
• Integration with Hermes agent orchestration layer

---

License

MIT — see LICENSE.

---

Acknowledgements

Built for the Hermes agent ecosystem and inspired by the Google Cloud Knowledge Catalog team's OKF draft specification. If you use Hermes or OKF, this library is designed to be your memory backbone.

⭐ Star this repo if you're building agent memory systems — let's make Hermes the best agent framework out there.