Metadata-Version: 2.4
Name: openmemo-openclaw
Version: 2.6.0
Summary: OpenMemo Cognitive Memory Adapter for OpenClaw Agents
Author-email: OpenMemo <hello@openmemo.ai>
License: MIT
Project-URL: Homepage, https://github.com/openmemoai/openmemo-openclaw-adapter
Project-URL: Documentation, https://openmemo.ai
Project-URL: Repository, https://github.com/openmemoai/openmemo-openclaw-adapter
Keywords: openmemo,openclaw,agent,memory,cognitive
Classifier: Development Status :: 4 - Beta
Classifier: Intended Audience :: Developers
Classifier: Programming Language :: Python :: 3
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Requires-Python: >=3.9
Description-Content-Type: text/markdown
Requires-Dist: openmemo>=0.23.1
Requires-Dist: requests>=2.28
Provides-Extra: yaml
Requires-Dist: pyyaml>=6.0; extra == "yaml"
Provides-Extra: dev
Requires-Dist: pytest>=7.0; extra == "dev"
Requires-Dist: pytest-asyncio>=0.21; extra == "dev"

# OpenMemo × OpenClaw Adapter

**Turn OpenClaw Agents into Persistent, Memory-Driven Collaborators**

Most AI agents today are stateless.

They forget:
- what tasks they already completed
- what decisions were made
- what tools worked before
- what the user prefers

So they repeat work. They redo the same tasks. They waste tokens and time.

**OpenMemo fixes this.**

OpenMemo adds a persistent memory layer to OpenClaw that allows agents to:
- remember past actions
- reuse successful workflows
- avoid duplicate work
- accumulate long-term knowledge

After installing OpenMemo, your OpenClaw agents stop behaving like stateless chatbots and start behaving like **long-term collaborators**.

---

## What OpenMemo Adds to OpenClaw

Installing the adapter enables:

### Persistent Memory

Agents remember:
- preferences
- past tasks
- decisions
- workflows
- observations

Example memory:

```
User switched from Flask to FastAPI
Scene: coding
Type: decision
Confidence: 0.92
```

### Task Deduplication

Before executing a task, OpenMemo checks whether it has already been done.

```
task: deploy backend
fingerprint: deployment|backend|docker-compose
status: success
```

If the task was already completed, the agent can **reuse**, **skip**, or **continue** — instead of repeating work.

### Scene-Aware Memory

OpenMemo detects the current working scene:
- `coding`
- `debugging`
- `research`
- `deployment`
- `analysis`

and retrieves only the most relevant memories.

### Structured Knowledge

OpenMemo stores atomic memory cells, not raw chat logs.

Instead of:
> long messy conversation history

it stores:

```
"Backend deployed using Docker Compose"
Scene: deployment
Type: task_execution
```

This dramatically improves recall quality.

### Automatic Memory

OpenMemo automatically extracts memory from:
- conversations
- tool calls
- task completions
- decision changes

No manual tagging required.

### Memory Inspector

OpenMemo includes a visual dashboard where you can see:
- what the agent remembers
- how memory is ranked
- how much token cost is saved

---

## Why This Matters

Modern agent frameworks like OpenClaw, LangGraph, CrewAI, and AutoGPT are enabling developers to build increasingly complex AI workflows.

But most agents still rely on short-term chat context. That means:
- no long-term learning
- no task reuse
- no accumulated experience

**OpenMemo introduces a memory infrastructure layer designed specifically for agents.**

---

## Architecture

OpenMemo sits between the agent and the LLM.

```
User / Agent Interaction
        │
        ▼
   Memory Capture
   (conversation, tool calls, tasks)
        │
        ▼
  OpenMemo Memory Engine
        │
  ┌─────┼─────┐
  │     │     │
Recall Ranking Dedup
  │     │     │
  └─────┴─────┘
        │
        ▼
  Context Injection
        │
        ▼
   LLM Reasoning
```

### Local-First Design

OpenMemo is local-first by default. All memory operations run locally.

```
Agent
  ↓
OpenMemo Local Server
  ↓
Memory Store
```

Cloud services are optional.

Benefits:
- **Privacy** — all data stays local
- **Low latency** — no network round trips
- **No external dependencies** — works offline

---

## Automatic Memory Rules

Installing OpenMemo also activates **Memory Rules**.

These rules guide the agent to properly use memory:

1. Prefer OpenMemo memory over raw conversation history.
2. Before executing any task, check if the task was already completed.
3. If a task already succeeded, reuse or skip it.
4. After meaningful task completion, write structured memory back to OpenMemo.

These rules are injected through a **merged soul context**, ensuring consistent memory behavior.

---

## Memory Inspector

OpenMemo includes a built-in **Memory Inspector Dashboard**.

The inspector shows:

| Metric | Description |
|--------|-------------|
| Memory Cells | Total structured memories stored |
| Tokens Saved | Token cost reduction from memory reuse |
| Recall Confidence | Quality score of memory retrieval |
| System Status | Health of memory infrastructure |
| Scene Dynamics | Current detected scene and confidence |

Users can override the detected scene if needed.

### Memory Stream

Real-time memory entries with ranking:

```
User switched from Flask to FastAPI
Scene: coding
Rank Score: 0.92
```

This makes the memory system **transparent and explainable**.

---

## Example Workflows

### Coding Agent

**Without OpenMemo:**

```
User: deploy backend
Agent: builds image → deploys

(later...)

User: deploy backend again
Agent: repeats everything from scratch
```

**With OpenMemo:**

```
User: deploy backend
Agent: deploys using Docker Compose → memory stored

(later...)

User: deploy backend again
Agent: detects prior deployment → reuses workflow
```

### Research Agent

Agent remembers:
> User prefers academic papers over blog posts

Future research queries prioritize papers automatically.

### DevOps Agent

Agent remembers:
> Production deployments use Kubernetes

Future deployment tasks follow the correct workflow automatically.

---

## Multi-Agent Memory (Upcoming)

OpenMemo also enables collaborative memory for multi-agent systems.

Memory is organized into three layers:

### Private Memory

Per-agent knowledge.

```
planner_agent/private
coder_agent/private
```

### Shared Task Memory

Agents working on the same task share memory.

```
task_123/shared
```

Stores: task goals, completed steps, execution results.

### Team Knowledge Memory

Long-term shared knowledge.

```
team/default
```

Stores: project conventions, reusable workflows.

---

## What Makes OpenMemo Different

| Feature | Chat History | RAG | **OpenMemo** |
|---------|-------------|-----|-------------|
| Persistent Memory | ❌ | ❌ | ✅ |
| Task Deduplication | ❌ | ❌ | ✅ |
| Scene Awareness | ❌ | ❌ | ✅ |
| Decision Tracking | ❌ | ❌ | ✅ |
| Memory Inspector | ❌ | ❌ | ✅ |

**OpenMemo is not just memory storage. It is agent memory infrastructure.**

---

## Installation

Install the adapter:

```bash
pip install openmemo-openclaw
```

Start OpenMemo:

```bash
openmemo serve
```

Run OpenClaw normally. Memory and recall will start automatically.

---

## Vision

We believe future software will be built around persistent AI agents.

These agents need:
- long-term memory
- shared knowledge
- explainable reasoning
- reusable workflows

**OpenMemo is building the memory infrastructure for AI agents.**

---

## Contributing

We welcome contributions from the community.

If you're interested in building the future of agent memory infrastructure, join us.

---

## License

MIT

---

**Built with [OpenMemo](https://github.com/openmemoai/openmemo) — The Memory Infrastructure for AI Agents.**
