Skip to main content
Identity files are the foundation of your agent’s memory. They’re automatically loaded into every interaction, giving your agent consistent personality, purpose, and awareness of who you are.

SOUL.md

SOUL.md defines your agent’s personality and communication style. It controls how your agent presents itself and interacts with you. By default, a Dialog agent uses the name “Dialog” with a professional but approachable tone. SOUL.md contains the agent name and emoji, a short tagline, persona description and tone guidelines, and behavioral boundaries the agent must follow. Behavioral boundaries are especially important. They ensure your agent stays grounded in real data — for example, always citing sources with full URLs, never fabricating data, and flagging low-confidence findings.

PURPOSE.md

PURPOSE.md defines your agent’s mission, the topics it focuses on, and how it measures success. When you first create an agent, PURPOSE.md starts with initialized: false. During onboarding, Dialog walks you through setting up the agent’s purpose by asking about your research goals and areas of focus.

What PURPOSE.md contains

  • Mission statement - A clear description of what the agent is built to do
  • Scope - The domains, topics, and industries the agent should focus on
  • Success criteria - What good output looks like for your use case
A well-defined purpose makes your agent significantly more useful. The more specific your scope and success criteria, the more focused and relevant your research results will be.

TASKS.md

TASKS.md is your agent’s persistent task list. It lives alongside the other identity files at the root of your workspace and is auto-loaded into every interaction. When you say things like “Add to my task list: review competitor pricing pages” or “What’s currently on my list?”, the agent reads and updates TASKS.md directly. It persists across sessions, so nothing falls through the cracks between conversations. You can also open the file from the workspace panel in the sidebar to review or edit it manually.

USER.md

USER.md stores facts about you so the agent can personalize its responses. Like PURPOSE.md, it starts with initialized: false and gets populated during your first conversation. Your agent learns about you through natural questions rather than a form. Over time, USER.md may include:
  • Your role and title
  • Your industry and company
  • Research preferences and priorities
  • Your timezone
The agent references this context on every interaction. If you’re a product manager in fintech, your agent will frame research findings through that lens without you having to remind it. All three identity files are auto-loaded for every interaction, including chat sessions and scheduled tasks. Whether you’re chatting at 9 AM or a scheduled task runs at midnight, the agent operates with the same identity and context.

Customizing identity files

You can edit identity files directly in your agent’s workspace to tune behavior. Common customizations include:
  • Changing the agent’s tone in SOUL.md to be more formal, more casual, or domain-specific
  • Narrowing the scope in PURPOSE.md to focus the agent on a specific market or competitor set
  • Updating your details in USER.md if your role or priorities change
Removing behavioral boundaries from SOUL.md (like source citation requirements) may result in less reliable research output. Modify boundaries with care.