Skip to main content
Dialog agents use a three-tier memory system that provides continuity across conversations, scheduled tasks, and devices.

​
The three tiers of memory

Every agent has three layers of memory, each serving a different purpose. Together, they ensure your agent always knows who it is, what it’s working on, and what you’ve discussed.
TierWhat it storesLoaded whenPersists across
Identity FilesAgent personality, mission, and user context (SOUL.md, PURPOSE.md, USER.md)Every interaction (auto-loaded)All sessions and scheduled runs
Workspace FilesTask lists, research notes, skills, and databasesOn demand, when the agent needs themAll sessions and scheduled runs
Session StorageChat messages, tool activity, and timestampsWhen you open a conversationAll devices where you’re logged in
Identity files define who your agent is. Workspace files store what your agent is working on. Session storage captures every conversation you’ve had.

​
Persistence guarantees

Your agent’s memory is tied to the agent itself, not to any individual chat session. This means:
  • Starting a new chat does not reset memory. Your agent still has access to all workspace files and identity context.
  • Scheduled tasks run with full access to the same workspace. A task that writes research notes at 2 AM is readable in your morning chat.
  • Cross-device access works automatically. Sign in on another device and your full chat history is there.
The total workspace size limit is 50MB shared across all files in your agent’s workspace, including identity files, research notes, and databases.

​
How memory enables continuity

Memory is what turns Dialog from a stateless chatbot into a persistent research partner. Here are a few ways it shows up in practice:
  • Personalized responses - Your agent reads USER.md on every interaction, so it knows your role, industry, and preferences without you repeating them.
  • Ongoing research - The agent saves notes to the research/ directory. When you come back days later and ask a follow-up, it picks up where it left off.
  • Accumulated knowledge - Databases in the databases/ directory let your agent track trends over time, building a richer picture with each research run.
  • Task management - TASKS.md keeps a running backlog, so your agent knows what’s been done and what’s still pending.

Identity Files

Your agent’s personality and context

Workspace Files

Persistent research, tasks, and data

Session Storage

Chat history and message persistence