Skip to main content

​
Core Concepts

​
Tasks

A task is a high-level instruction you give to Miteos. The platform decomposes it into subtasks and assigns agents. 
"Research competitor pricing and build a comparison dashboard"
→ Subtask 1: Research (researcher agent)
→ Subtask 2: Analyze (analyst agent)
→ Subtask 3: Build (builder agent)

​
Agents

Agents are specialized AI workers. Each has a role, tools, and execution strategy:
TypeRoleTools
ResearcherFind information from the webBrowser, search, extraction
BuilderCreate code, files, dashboardsCode execution, file system
AnalystProcess data, generate insightsCode execution, charts
TraderExecute market strategiesTrading engine, broker APIs
EmailManage inbox and outreachEmail APIs, drafting
SocialCreate and publish contentPlatform APIs, scheduling

​
War Room

The War Room is the real-time command center. It shows:
  • Agent Fleet Rail — All active agents with status, progress, tools in use
  • Workspace — Browser viewport, code editor, files, canvas, terminal
  • Timeline — Chronological log of every action across all agents
  • Deliverables — Completed outputs ready for download or deployment

​
Autonomy Dial

The autonomy dial (0-100) controls how much independence agents have:
  • 0-20 — Copilot mode: agent suggests, you confirm every action
  • 20-50 — Supervised: agent acts on low-risk, asks for high-risk
  • 50-80 — Autonomous: agent handles most actions, asks for critical ones
  • 80-100 — Full auto: agent executes everything, notifies after completion

​
Skills

Skills are reusable patterns extracted from successful task executions. When agents complete tasks well, the platform can compile the action sequence into a named skill that runs faster next time.

​
MCP Integrations

Miteos uses the Model Context Protocol (MCP) to connect external services as agent tools. When you connect Gmail, the agent gains tools like read_inbox, send_email, search. Each integration runs as an isolated container.

​
Workspaces

Workspaces are isolated environments where agents work. Each workspace has:
  • File storage (code, reports, data)
  • Browser sessions
  • Agent memory (context that persists across tasks)
  • Deliverables output

​
Memory

Three-layer memory system:
  1. Working — Current task context (cleared after task)
  2. Session — Per-workspace context (persists across tasks in same workspace)
  3. Persistent — Per-user long-term memory (learned preferences, patterns)