Skip to main content
Dialog is built on a modular architecture where agents, memory, tools, skills, channels, and scheduling work together to power your AI workforce. This page provides a high-level map of how these systems connect.

​
How it all fits together

Agents

Specialized AI assistants with their own workspace, tools, and personality

Memory & Workspace

Persistent storage that gives your agent continuity and a growing knowledge base

Tools

74+ data sources and integrations your agent uses

Skills

Reusable custom workflows you teach your agent

Scheduling

Automated recurring research, monitoring, and briefings

Channels

Where results get delivered (Telegram, Slack, Discord, WhatsApp)

​
The interaction flow

Every interaction follows the same path through Dialog’s systems, whether it’s a research question, a link to save, or a task to add.
1

You send a message

Type anything natural — a research question, a URL to save, a task to add, a content draft request, or a follow-up to previous work.
2

Agent loads context

Your agent loads its identity files (PURPOSE.md, SOUL.md, USER.md), workspace state, and session history. This gives it full context on who you are, what it’s been working on, and how to help.
3

Agent selects tools

Based on your message, the agent determines which tools to use. A research question routes to Reddit and web search. A URL triggers web scraping. A calendar request hits your connected Google Calendar.
4

Tools execute

The selected tools run in parallel where possible — fetching posts, scraping pages, pulling social data, or querying connected apps. Each tool returns structured results.
5

Agent processes and responds

The agent synthesizes results, updates workspace files if needed (saving links, updating tasks, filing research), and streams the response back to you in real time.

​
Real-world workflows

Here’s how the systems come together for common use cases:

​
Competitive intelligence pipeline

  1. Save competitors — Send URLs as you discover them. Agent scrapes, enriches, and files each one.
  2. Research — Ask the agent to analyze positioning, pricing, and features across your saved competitors.
  3. Track changes — Set up scheduled monitoring to flag competitor updates weekly.
  4. Brief — Get a synthesized competitive brief delivered to Telegram every Monday.

​
Content creation workflow

  1. Capture ideas — Send content ideas to your agent as they come to you.
  2. Research — Ask the agent to explore a topic across Reddit and the web for inspiration.
  3. Draft — Request post drafts with specific angles and audiences.
  4. Iterate — Workshop the copy through back-and-forth until it’s ready.

​
Daily operations

  1. Morning briefing — Scheduled task delivers overnight signals and task summary.
  2. Throughout the day — Save links, add tasks, ask questions, file research.
  3. End of day — Agent produces a recap of everything captured and completed.

​
System boundaries

Each agent operates within its own isolated environment. Agent isolation. Every agent has its own persistent workspace with dedicated storage, tool configuration, and skills. One agent cannot access another agent’s workspace, files, or conversation history. Tool configuration is per-agent. Each agent can have different tool preferences — one agent might have social media sources enabled while another only uses Reddit and web search. Scheduled tasks share the workspace. When you set up scheduled research, those tasks run in the same workspace as your chat sessions. This means scheduled tasks can read and write the same files, use the same skills, and build on the same memory from interactive conversations.
Agents are fully independent by design. If you need to share research between agents, use connected apps to store findings in shared tools like Notion or Google Docs.