Skip to main content
Your agent’s workspace is a persistent file system where it stores everything it needs to be useful across sessions — research notes, task lists, identity files, bookmarks, custom skills, and structured databases.

Workspace structure

Every agent workspace follows a consistent directory structure:

Identity files

These files define who your agent is and how it works with you:
  • PURPOSE.md — Describes the agent’s mission and focus area. A Chief of Staff’s purpose is different from a Social Media Manager’s or a Product Analytics Agent’s.
  • SOUL.md — Defines the agent’s personality, tone, and behavioral guidelines. This shapes how the agent communicates and approaches tasks.
  • USER.md — Stores information about you that the agent uses to personalize its work — your role, preferences, ongoing projects, and context.
These files are loaded at the start of every session, giving your agent consistent context. You can edit them directly by telling your agent:
For more on identity files, see Identity Files.

TASKS.md

TASKS.md is your agent’s task backlog. It tracks what the agent is working on, what’s queued up, and what’s been completed. The file is organized into three sections:
  • Active — Tasks currently in progress
  • Backlog — Tasks queued for future work
  • Completed — Finished tasks with results or summaries
Your agent updates TASKS.md as it works. When you ask it to research something or set up a scheduled task, it adds an entry. When the work is done, it moves the item to Completed with a summary. For more on task management, see Managing Tasks.

research/

The research/ directory is where your agent saves research notes, analysis, competitive briefs, and collected bookmarks. When Dialog completes a research task, it persists the findings here so you can reference them later. This is especially useful for multi-session research. Your agent can pick up a previous analysis, compare new findings against earlier results, or build on past work without starting from scratch. Common patterns:
  • Competitive research files — Ongoing competitive landscape documents that grow as you save more competitor info
  • Topic collections — Grouped links and notes around a specific theme
  • Bookmark files — Saved URLs with summaries and tags

skills/

The skills/ directory holds custom skills that extend what your agent can do. Skills placed here are auto-discovered, meaning your agent can find and use them without extra configuration. For details on creating and managing skills, see Skills.

databases/

The databases/ directory stores SQLite databases for structured data. This lets your agent track and query data over time.

Working with databases

  • Your agent accesses databases via command line tools
  • Multi-statement writes use transactions to keep data consistent
  • Database files follow a lowercase-with-hyphens naming convention (e.g., competitive-intel.db)

Example use cases

  • Bookmark management — Store saved links with enrichment fields, tags, and metadata
  • Content pipeline — Track content ideas from ideation through drafting to publication
  • Competitive tracking — Build a structured dataset of competitor mentions, features, and pricing
  • Performance metrics — Store analytics data (LinkedIn impressions, engagement) for trend analysis
  • Research indexing — Catalog sources and findings across sessions for fast lookup
Databases are ideal when you need to compare data across time periods. If your agent runs a weekly competitive scan, it can append results to the same database and spot changes automatically.

Workspace limits

All files in your agent’s workspace share a 50MB total size limit.
If your workspace approaches the 50MB limit, consider archiving older research files or cleaning up completed task entries. Large databases are usually the first place to look.