#Memory

SciLaxy agents can remember information across conversations. The memory system has two layers: core memory for structured, always-available context, and long-term memory for searchable item-based storage.

#Core Memory

Core memory is a set of structured sections that are always included in the agent's system prompt. This gives the agent persistent context about the user and the current working state.

Core memory is small and focused — it's the information the agent should always have at hand.

#Memory Sections

Each agent has four core memory sections:

Agents can read and update these sections during conversations. For example, if a user says "I prefer concise responses," the agent can update the Preferences section to remember this.

#Viewing and Editing Memory

You can view and edit an agent's core memory through the web interface:

1. Open the agent settings
2. Navigate to the Memory section
3. View or edit each memory section

#Long-Term Memory

Beyond core memory, agents have access to a long-term memory store backed by LangGraph Store. This is a searchable collection of individual memory items.

Long-term memory is useful for:

• Storing facts learned during conversations
• Remembering user corrections ("Actually, our API uses v2, not v1")
• Accumulating knowledge over time

Agents can search, create, update, and delete memory items using the memory tools.

#How Agents Learn

Agents learn from conversations through two mechanisms:

1. Explicit updates — When the agent recognizes important information, it writes to core memory or creates a long-term memory item
2. User corrections — When a user corrects the agent, it can update its memory to avoid repeating the mistake

Memory persists across sessions. The next time the user starts a conversation, the agent already knows what it learned previously.