six6 is a zero-dependency, file-based memory and cognition toolkit for AI agents (built for OpenClaw). Six independent Python modules give an agent short- and long-term memory, nightly reflection, daydream-style idea generation, an idea lifecycle "topic lab", an inbox task loop, and a cron heartbeat — all communicating only through Markdown and JSONL files.

中文：six6 是一套给 AI Agent 用的文件系统记忆与认知模块（OpenClaw Skill），6 个模块互不依赖，只靠 Markdown / JSONL 文件交换状态，无需向量数据库、无需第三方依赖。

🎯 Who is it for

适用场景:

• You want to add persistent AI agent memory (long-term, file-based) to OpenClaw or your own agent.
• You prefer human-readable, git-diffable Markdown memory over a vector store.
• You want low-cost "background thinking" (reflection, daydreaming) driven by cron or systemd.
• You only need one module and want to take it on its own.

📋 Requirements

• Python 3 (CI runs 3.11); the scripts use only the standard library.
• Optional: the gh CLI — only needed for topic-lab planting and for labelling escalations.
• Optional: an OpenAI-compatible or Anthropic-compatible LLM API — only needed by meditation and daydream. Configure it with LLM_API_BASE, LLM_API_KEY, LLM_MODEL and LLM_API_TYPE (see skill-meditation).

🚀 Quick Start

# 1. Clone
git clone https://github.com/ythx-101/six6.git
cd six6

# 2. Install: creates the base dir and copies scheduling templates into it
bash distribution/install.sh --base-dir ~/agent

# 3. Validate the protocol files
python3 runtime/scripts/six6.py validate --base-dir ~/agent

# 4. Fire a heartbeat pulse
python3 runtime/scripts/six6.py pulse heartbeat --base-dir ~/agent

Then schedule the pulses with distribution/templates/cron.example (a copy is placed in ~/agent/examples/) or distribution/templates/systemd.example. Use python3 runtime/scripts/six6.py doctor --base-dir ~/agent for a health check.

Try the demo

python3 runtime/scripts/six6.py validate --base-dir demo
python3 runtime/scripts/six6.py pulse heartbeat --base-dir demo

Note: some demo commands (such as doctor or farm.py --add-water) write run files into demo/. See demo/README.md.

🔁 Usage example

A full loop: remember → reflect → daydream → grow ideas.

BASE=~/agent
python3 skill-memory/scripts/remember.py "Read a paper on agent memory" --base-dir $BASE
python3 skill-memory/scripts/update_now.py "Drafting a post about file-based memory" --base-dir $BASE
# needs LLM_API_KEY:
python3 runtime/scripts/six6.py pulse nightly --base-dir $BASE   # meditation -> MEMORY.md
python3 runtime/scripts/six6.py pulse idle --base-dir $BASE      # daydream -> new seed
python3 runtime/scripts/six6.py pulse daily --base-dir $BASE     # topic-lab tick
python3 skill-topic-lab/scripts/farm.py --add-water <seed-id> --base-dir $BASE

The base directory created by six6.py init looks like this:

~/agent/
├── NOW.md                      # short-term context
├── MEMORY.md                   # long-term core memory
├── memory/                     # daily episodic notes (YYYY-MM-DD.md)
├── data/
│   ├── inbox.jsonl             # task / event queue
│   ├── topic-lab-seeds.jsonl   # idea seeds
│   ├── evolution.md            # meditation output log
│   └── health.json             # written by `six6.py doctor`
└── deadletter/                 # failed inbox items

🆚 Comparison

A qualitative comparison of approaches (not a benchmark):

😤 Problem

You: "I want an AI Agent that can think, remember, and generate ideas."
Current frameworks: "Here is a massive 1GB monolithic orchestrator with tight Python couplings."

six6 decomposes the core cognitive functions of an AI Agent into 6 independent modules. They do not import each other. They do not share memory space. They communicate entirely through a standardized set of files (Markdown & JSONL).

To make this reusable for OpenClaw users, the repository also includes three supporting layers that sit underneath the six modules:

• protocol/: file contracts and schemas.
• runtime/: init, validate, doctor, and pulse entrypoints.
• distribution/: packaging guidance for install and migration.

You can install just the memory system, or you can install the entire suite to create a living, breathing organic AI entity.

🧬 Architecture

Absolute decoupling is our core principle:

• No Python imports between modules.
• Standardized data flow via Markdown (for cognition/memory) and JSONL (for queues/seeds).

Read the detailed ARCHITECTURE.md for data flow specifications.

🧱 The 6 Modules

1. skill-memory (Cerebral Cortex)

   • Manages short-term context (NOW.md) and long-term daily episodic memory (memory/YYYY-MM-DD.md).
   • Maintains the core cognitive summary (MEMORY.md).

2. skill-meditation (Reflection)

   • Deep reflection mechanism. Runs nightly to synthesize the day's memory into the core MEMORY.md.

3. skill-daydream (Divergent Ideation)

   • Randomly samples historical memory fragments during idle time to generate serendipitous ideas, outputting them as JSON objects to the Topic Lab.

4. skill-topic-lab (Greenhouse / Farm)

   • Manages the lifecycle of generated ideas (seeds).
   • Tracks maturity. When a seed reaches maturity ≥ 80 on a daily tick, it automatically triggers external outputs (e.g., creating a GitHub Issue).

5. skill-autoloop (Action Center)

   • The task queuing and execution hub.
   • Manages inbox.jsonl and handles escalation protocols (e.g., needs-decision).

6. skill-monitor (Life Support / Heartbeat)

   • The chronometer and health checker. Manages system pulses and triggers for the other modules.

❓ FAQ

What is six6? (six6 是什么？)
A file-based memory and cognition toolkit for AI agents: six decoupled Python modules that share state only through Markdown and JSONL files.

Does six6 need a vector database?
No. Everything is stored as Markdown and JSONL. (six6 does not provide semantic search.)

Which LLMs does it work with?
Meditation and daydream work with any OpenAI-compatible /chat/completions endpoint or Anthropic-compatible /v1/messages endpoint; <think> blocks are stripped from replies. See skill-meditation for Ollama and MiniMax examples.

Can I use only the memory module? (可以只用记忆模块吗？)
Yes. Modules do not import each other.

Do I have to use OpenClaw?
No. The project is packaged for OpenClaw, but the scripts do not depend on it; any agent or cron job that can run Python scripts and read files can use them.

How do I schedule it?
Use the cron and systemd templates in distribution/templates/.

What happens when an idea matures?
On a daily tick, a seed with maturity ≥ 80 becomes a GitHub Issue (label idea) via gh and is marked planted. Without gh, planting is skipped.

How do I migrate from a clawd workspace?
Run distribution/migrate_from_clawd.py (see distribution/README.md); it only appends/merges.

Is my data sent anywhere?
Meditation and daydream send memory text to the LLM endpoint you configure. If gh is installed, topic-lab sends a mature seed's topic and description to GitHub as an Issue, and autoloop adds a needs-decision label to escalated Issues. Nothing else leaves your machine.

📄 License

MIT License. See LICENSE for details.