Deepmiro
About
Simulate hundreds of AI agents to predict how communities react to events and policies. Upload any document (PDF, Markdown, text) and DeepMiro spawns a diverse swarm of AI agents that debate, share, and form opinions — then delivers a calibrated prediction report. Free and open-s
Details
- Author
- kakarot-dev
- Downloads
- 311
- Categories
- Other
Jump to
- Open‑source under AGPL-3.0 license.
- Self‑hosted via Docker Compose.
- Hosted option available at deepmiro.org.
- Fully offline with Ollama support.
- Supports upload of PDF, Markdown, and text files.
- Allows interviewing any simulated agent post‑simulation.
Setting up with Highlight
This MCP is not yet compatible with Highlight’s one-click setup. However, you can still use it with Highlight by following these steps:
- Download and install Highlight from highlightai.com/download
- Navigate to the plugins tab and select "Add Custom Plugin"
-
Configure the plugin with the settings below
Plugin Name
DeepmiroCommand (node, npx, python, etc.)Please refer to the README for specific instructions on how to obtain API keys or other required environment variables.
- Enable "Start Automatically" if you want the plugin to start when Highlight launches
From the repository
DeepMiro can be self-hosted via Docker Compose or used through the hosted option at deepmiro.org. After installation or sign‑up, use the provided tools (e.g., create_simulation, simulation_status, get_report, interview_agent) to start predictions, monitor progress, fetch reports, or chat with individual agents.
create_simulation
Run a swarm prediction — graph build, persona generation, multi-agent simulation, report. IMPORTANT: Enrich the prompt before calling. The engine extracts named entities to create personas. Add specific people, companies, organizations, and opposing viewpoints. Show the enriched prompt to the user for confirmation first. If the user provides a document (PDF, MD, TXT), call upload_document first and pass the returned document_id. Returns immediately with simulation_id. Call simulation_status to wait for completion — each call blocks up to 50s for the next state change, so you only need a few. When status returns state=COMPLETED, the full report is included inline.
simulation_status
Check the progress of a running or completed simulation. Long-polls by default — blocks up to 50s waiting for a state change (phase transition, new round, new actions, completion). When state=COMPLETED, includes the full prediction report inline. Lifecycle: CREATED → GRAPH_BUILDING → GENERATING_PROFILES → READY → SIMULATING → COMPLETED/FAILED/CANCELLED/INTERRUPTED.
get_report
Generate and retrieve the prediction report for a completed simulation. If the report hasn't been generated yet, triggers generation (may take 1-3 minutes). Returns a detailed markdown analysis ready to display as an artifact in the side panel. Pass force_regenerate=true to rebuild an already-cached report.
interview_agent
Chat with a specific simulated agent to understand their perspective, reasoning, and predicted behavior. The agent responds in character based on their persona and simulation experience.
list_simulations
List past simulation runs with their status and metadata.
search_simulations
Search past simulations by topic, project name, or simulation ID.
upload_document
Upload a document for use in simulations. LIMITS: Max 10MB, PDF/MD/TXT only. The server extracts text server-side (PyMuPDF for PDFs). Returns a document_id to pass to create_simulation. NOTE: Only works with local file paths (stdio transport). For remote/hosted mode, the client skill uploads via HTTP instead.
simulation_data
Access simulation data: agent profiles, configuration, action logs, social media posts, round-by-round timeline, per-agent activity stats, and interview history. Paginated — use offset to get more results when has_more is true.
cancel_simulation
Stop a running simulation. SIGTERMs the subprocess immediately and marks the simulation as stopped. Partial action log is preserved — you can still call get_report or simulation_data on a cancelled simulation for whatever data was produced before cancellation. Use this when a simulation is taking too long, was started by mistake, or is producing bad output you want to abort.
Claude Desktop / Cursor
Paste into your MCP client config file to install this server.
{
"mcpServers": {
"deepmiro": {
"deepmiro": {
"command": "npx",
"args": [
"-y",
"deepmiro-mcp"
],
"env": {
"DEEPMIRO_API_KEY": "<YOUR_API_KEY>"
}
}
}
}
}McpServers
{
"deepmiro": {
"command": "npx",
"args": [
"-y",
"deepmiro-mcp"
],
"env": {
"DEEPMIRO_API_KEY": "<YOUR_API_KEY>"
}
}
}DeepMiro
Simulate hundreds of AI agents to predict how communities react to events and policies. Upload any document (PDF, Markdown, text) and DeepMiro spawns a diverse swarm of AI agents that debate, share, and form opinions — then delivers a calibrated prediction report. ## Tools - create_simulation — Start a new prediction from a prompt or uploaded document - simulation_status — Check progress of a running simulation - get_report — Fetch the final prediction report - interview_agent — Chat with any simulated agent post-simulation - list_simulations — View past simulation runs - search_simulations — Find simulations by topic - quick_predict — Fast, lightweight prediction (sync) ## Features - 🔓 Open-source (AGPL-3.0) - 🐳 Self-hosted via Docker Compose - ☁️ Hosted option at deepmiro.org - 🦙 Ollama support for fully offline use ## Links - GitHub: https://github.com/kakarot-dev/deepmiro - Website: https://deepmiro.orgSign in to leave a review
Use Google, GitHub, or an email account so ratings stay tied to real people.
No reviews posted yet.
