Memory system of the MFO Server¶

The MFO Server provides a memory system designed as a key-value store, which tasks and AI agents can use to store and retrieve contextual data dynamically. Here’s how it works and how it can be leveraged:

🔑 Memory: Core Concept¶

• The memory system allows user-scoped data storage. Each memory entry is identified by:

• user_id: isolates memory between different users or agents.

• key: the unique identifier for a piece of memory.

• value: any JSON-serializable data.

• Use case: AI agents can save intermediate results, user preferences, session states, or context variables across multiple tasks, making conversations or workflows stateful and adaptive.

🛠️ API & Handlers¶

• POST /api/memory: Set a key-value pair.
• GET /api/memory/\:key: Retrieve the value of a key.
• PUT /api/memory/\:key: Update a value.
• DELETE /api/memory/\:key: Remove a key.

Example: When an AI agent receives a question like “What’s my last order status?”, it can fetch a stored key like "last_order_status" from memory.

👷 How Tasks & Agents Use It¶

1️⃣ Temporary State: A long-running workflow (e.g., processing a customer complaint) can save its current progress (e.g., "complaint_status": "investigating").

2️⃣ Cross-Task Sharing: One task stores data (Set), and another retrieves it later (Get). This enables modular task orchestration.

3️⃣ Memory Batching: Agents can process multiple memory operations at once using the BatchProcessor, which allows bulk get/set/delete to optimize performance.

🗂️ Under the Hood¶

• PostgreSQL-backed memory store:

• Table: slop_memory

• Columns: user_id, key_name, value (JSONB), expires_at

• Indexed for fast retrieval and with expiration logic to automatically clean up stale data.

• In-Memory Option: For testing or fast tasks, there’s an InMemoryStore implementation based on Go’s sync.RWMutex and in-memory maps.

🧠 Practical Scenarios¶

• Conversation Memory: A chatbot keeps track of recent user inputs to make follow-up replies smarter.

• Pipeline Tracking: A multi-step job tracks its execution state in memory to resume after a restart.

• AI Agent Coordination: Two agents working on related tasks (e.g., parsing and summarizing emails) can use memory to share interim results seamlessly.

✅ Benefits¶

In summary: The memory system in the MFO framework is like the short-term working memory of the brain: it gives agents and tasks a place to temporarily store knowledge that can be reused to create seamless, context-aware workflows.