Memory¶

Vandelay's memory system gives your agent persistent context across conversations, sessions, and restarts.

Memory Layers¶

1. Session History¶

Agno stores conversation history per session in the SQLite database. Configured with:

• num_history_runs=2: Include the last 2 conversation turns for context
• max_tool_calls_from_history=5: Limit tool call results in history to manage tokens
• enable_session_summaries=True: Summarize older sessions for long-term context

2. Agentic Memory¶

Agno's built-in memory system (update_memory_on_run=True) automatically extracts and stores important facts, preferences, and context from conversations. These are recalled in future sessions.

3. Workspace Files¶

The agent's workspace (~/.vandelay/workspace/) contains markdown files that shape its behavior and context. These are loaded into the system prompt at startup.

See Workspace Files for the full reference.

Storage¶

All memory is stored in a unified SQLite database at ~/.vandelay/data/vandelay.db. This handles:

• Session history
• Agentic memory records
• Session summaries

For PostgreSQL, set db_url in config or the DATABASE_URL environment variable.

Shared Memory Across Channels¶

All channels (Terminal, Telegram, WhatsApp, WebSocket) use the same user_id, so memory is shared. A conversation in Telegram is remembered when you switch to Terminal.

Token Management¶

History can grow large. Vandelay manages this with:

• Limited history runs (2 by default)
• Capped tool call results from history (5 by default)
• Session summaries that compress old conversations