Skip to main content
The Miteos War Room is your mission control for AI agents — you describe a goal, agents spin up and get to work, and you collect the results. This guide walks you through submitting your first task from scratch, watching your agents execute it in real time, and pulling down your finished deliverables.

Prerequisites


Running Your First Task

1

Open the War Room

From the Miteos dashboard, click New Task in the top-right corner. This opens the War Room — your real-time command center where you can watch agents work, review their actions, and collect deliverables.The War Room has four main panels:
  • Agent Fleet Rail — lists every active agent with its current status and tool in use
  • Workspace — a live browser viewport, code editor, file system, and terminal your agents control
  • Timeline — a chronological log of every action taken across all agents
  • Deliverables — completed outputs ready to download or deploy
2

Write your task prompt

In the Task Prompt field, describe what you want the agents to accomplish. Think of it like briefing a skilled team — the clearer you are about the goal and the expected output, the better the result.Good prompt examples:
  • “Research the top 5 AI writing tools competing with us, compare their pricing pages, and produce a structured comparison table as a CSV and a one-page PDF summary.”
  • “Audit our Google Ads account and identify campaigns with a CTR below 1%. List each campaign, its current spend, and a recommended fix for each.”
  • “Write a 60-second product demo video script for our project management app, then generate the video in 16:9 format with animated slides and a voiceover.”
You can always send guidance mid-task if you want to steer the agents after they’ve started. See Human-in-the-Loop for details.
3

Configure your task

Click Configure (below the prompt field) to expand the task settings:
SettingWhat it doesRecommended starting value
Max AgentsHow many parallel agents can run at once (1–10). More agents = faster completion on complex tasks.2–3
Autonomy LevelA 0–100 dial controlling how independently agents act. At 0 they ask before every action; at 100 they execute and notify afterward.70 for most tasks
Auto DeployIf agents build a website or app, deploy it automatically when the task finishes.Off until you’ve connected a deploy target
Deploy TargetWhere to publish if auto-deploy is on: Vercel, Netlify, Cloudflare Pages, or Expo.
For your first task, leave Autonomy Level around 70 and Max Agents at 2. This gives agents enough freedom to move fast while still pausing for anything sensitive.
4

Launch and monitor agent execution

Click Run Task. Miteos decomposes your prompt into subtasks, spins up the right agents, and assigns each one a role (Researcher, Builder, Analyst, etc.).In the Agent Fleet Rail you’ll see each agent card update in real time:
  • 🟡 Thinking — the agent is planning its next action
  • Acting — the agent is using a tool (browsing a URL, running code, reading a file)
  • Done — the agent has completed its assigned subtask
The Workspace panel shows exactly what each agent is looking at — a live browser, code being written, files being saved. The Timeline logs every action with a timestamp so you can audit the full run later.
Complex tasks with multiple agents typically complete in 2–10 minutes. You can close the browser tab and come back — agents keep running in the background and you’ll receive a notification when the task finishes.
5

Review and download your deliverables

When all agents finish, the Deliverables panel lights up in the bottom-right of the War Room. Each output (PDF, CSV, code bundle, video file, etc.) appears as a named card with its file type.
  • Click Download on any deliverable to save it locally.
  • If agents built a deployable app or site, a Deploy button appears — click it to push straight to your connected deploy target.
  • Click the deliverable name to preview it inline (supports PDF, Markdown, images, and HTML).
All deliverables are also stored in your workspace at platform.miteos.com and accessible via the API. They don’t expire.

Writing Effective Prompts

The single biggest factor in task quality is prompt quality. Vague prompts produce generic results; specific prompts with a clear output format produce work you can use immediately.
Five rules for great Miteos prompts:
  1. State the output format explicitly. “…and export the results as a CSV with columns: Company, Price, Features, Pros, Cons.”
  2. Set scope boundaries. “Focus only on US-based companies with under 500 employees.”
  3. Mention the audience. “Write this for a non-technical founder, avoid jargon.”
  4. Specify length or depth. “A 2-page executive summary, not a full report.”
  5. Include a success criterion. “The comparison should be good enough to present to our board.”
Vague vs. specific — a quick comparison:
❌ Vague✅ Specific
”Research our competitors""Research Notion, Coda, and Confluence. For each: current pricing tiers, G2 rating, top 3 user complaints from reviews. Output as a Markdown table."
"Write a marketing email""Write a 200-word re-engagement email for SaaS users who haven’t logged in for 30 days. Subject line should create urgency. CTA: restart free trial."
"Make a video""Create a 45-second 9:16 product demo video for TikTok showing our mobile app’s onboarding flow. Upbeat background music, on-screen captions.”

Task Examples

These real-world prompts show the range of work Miteos can handle across different verticals: Market research
“Scan Reddit, G2, and Trustpilot for the top 10 complaints users have about [competitor]. Categorize them by theme and rank by frequency. Output: a PDF report with charts.”
Ads audit
“Connect to my Google Ads account and run a full audit. Flag any ad groups with a Quality Score below 5, campaigns overspending their daily budget, and ad copy with no call-to-action. Generate a prioritized PDF fix list.”
Content production
“Draft a 4-week LinkedIn content calendar for a B2B fintech startup. 3 posts per week: 1 thought leadership, 1 product tip, 1 industry news reaction. Include suggested hashtags and post times.”
Lead generation
“Find 50 Series A SaaS companies in the HR tech space that raised funding in the last 6 months. Score each against our ICP (50–200 employees, US-based, no existing ATS). Output: scored CSV with LinkedIn URL and founding year.”
App build + deploy
“Build a pricing comparison landing page for our SaaS product. Use our brand colors (#1A1A2E, #E94560). Include a sticky nav, feature comparison table, and FAQ accordion. Deploy to Vercel when done.”