Skip to main content
When you give Miteos a goal, the platform doesn’t hand it to a single AI and hope for the best. Instead, an orchestrator breaks your instruction down into a directed graph of subtasks and dispatches each one to the right specialized agent — all running in parallel, coordinating until your deliverable is ready.

​
Tasks

A task is any high-level instruction you submit to Miteos. It can be a single sentence or a detailed brief — the orchestrator figures out the rest. Under the hood, Miteos parses your prompt, plans a dependency graph of subtasks (a DAG), and assigns each node to the most suitable agent type. Example decomposition: 
Prompt: "Research competitor pricing and build a comparison dashboard"
│
├── Subtask 1 → Researcher agent
│     Find pricing pages for all major competitors
│     Extract plan names, prices, and feature lists
│
├── Subtask 2 → Analyst agent  (runs after Subtask 1)
│     Clean and normalize the extracted data
│     Identify pricing gaps and positioning opportunities
│
└── Subtask 3 → Builder agent  (runs after Subtask 2)
      Scaffold a React dashboard
      Populate with analyzed data
      Deploy to preview URL
Subtasks without dependencies run in parallel, so a task that would take hours sequentially often completes in minutes.

​
Agent Types

Each agent type ships with a fixed set of tools purpose-built for its domain. Miteos routes subtasks to the agent whose toolset best matches the work required.
Agent TypeRoleAvailable Tools
ResearcherFind and extract information from the webStealth browser, web search, page extraction, screenshot
BuilderWrite, run, and deploy code and filesCode execution, file system, package manager, deploy targets
AnalystProcess data and generate insightsData processing, statistical analysis, chart generation
TraderExecute and backtest market strategiesBroker APIs, backtesting engine, signal scanning, Pine Script export
EmailManage inboxes and run outreachIMAP, Gmail, Outlook, draft creation, send (with approval)
SocialCreate and publish content across platformsPlatform APIs, content scheduling, analytics

​
Task Lifecycle

Every task moves through the same five-stage lifecycle, visible in real time inside the War Room.
1

Submit your prompt

Enter your instruction in the task input on the dashboard or via the API. Attach any relevant files or workspace context before submitting.
2

Orchestrator decomposes

The orchestrator parses your prompt, generates a DAG of subtasks, selects agent types for each node, and queues everything for execution.
3

Agents run in parallel

Independent subtasks spin up simultaneously. Each agent works inside an isolated workspace with its own browser session, file storage, and memory. You can watch them live in the War Room.
4

Results collected

As each subtask completes, its output is passed downstream to any dependent subtasks. The orchestrator assembles partial results into a coherent whole.
5

Deliverables ready

Final outputs — reports, dashboards, code, drafted emails, or deployed URLs — appear in the Deliverables panel and are saved to your workspace for download or further action.

​
Task Configuration

When you create a task, you can tune several settings that control execution behavior. These can be set in the task creation UI or passed as fields in the API request body.
OptionDefaultDescription
max_agents3Maximum number of agents that can run concurrently for this task
autonomy_level70Agent independence on a 0–100 scale — see Autonomy Dial
auto_deployfalseAutomatically deploy any code artifacts the Builder agent produces
deploy_targetvercelDeployment platform when auto_deploy is enabled — vercel, netlify, cloudflare, or expo
You can override autonomy_level mid-task using the slider in the War Room without cancelling or resubmitting the task.