Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/UnkleFunk/HouseMusicSwarm-/llms.txt

Use this file to discover all available pages before exploring further.

OpenSwarm is a fully open-source, terminal-first multi-agent AI system built on Agency Swarm — the framework powering real AI agencies. Instead of a single model trying to do everything poorly, OpenSwarm gives you 10 specialized agents coordinated by a central orchestrator: Virtual Assistant, Deep Research, Data Analyst, Slides Agent, Docs Agent, Image Generation, Video Generation, Reference Finder, and Ableton Dream Agent. Send one prompt from your terminal and watch the team produce complete, polished deliverables — slide decks, research reports, data visualizations, documents, images, videos, and even Ableton Live sketches. OpenSwarm supports OpenAI, Anthropic, and Google Gemini as AI providers, and connects to 10,000+ external services via Composio.
OpenSwarm is built on Agency Swarm, the open-source framework for building production-grade multi-agent systems. All agents, communication flows, and tool definitions follow Agency Swarm conventions and are fully customizable.

​
How it works

OpenSwarm uses an orchestrator-specialist model. Every user request flows through the Orchestrator first — it never answers questions or executes tasks directly. Instead, it analyzes your prompt and routes work to the right specialist(s) using two communication primitives:
  • SendMessage — dispatches a task to one or more specialists in parallel. The orchestrator waits for all responses before synthesizing a final reply. Used when a request spans multiple domains (e.g., research + slides + a cover image).
  • Handoff — transfers control to a single specialist for focused, sequential work. Any agent can hand off to any other agent, enabling multi-step workflows that chain specialists together automatically.
The orchestrator is defined in swarm.py with SendMessage flows from the orchestrator to every specialist, and bidirectional Handoff flows between all agents. This topology means your prompt can trigger a cascade of coordinated specialist work — no manual routing required.

​
Agent roster

Orchestrator

Routes every user request to the right specialist(s). Never answers directly — pure coordination via SendMessage and Handoff.

Virtual Assistant

Handles everyday tasks: writing, scheduling, messaging, and task management. Gains 10,000+ external integrations via Composio (Gmail, Slack, GitHub, HubSpot, and more).

Deep Research

Conducts comprehensive, evidence-based web research with citations, Scholar search, and balanced analysis.

Data Analyst

Analyses structured data, builds charts, and runs statistical models — all inside an isolated IPython kernel with pandas, NumPy, SciPy, and Plotly.

Slides Agent

Generates complete, visually polished HTML slide decks and exports them to PPTX. Uses stock photo search via Pexels, Pixabay, and Unsplash.

Docs Agent

Creates formatted Word documents and PDFs from outlines or raw content. Supports DOCX, PDF, Markdown, and TXT output.

Image Generation

Generates and edits images using Gemini 2.5 Flash Image, Gemini 3 Pro Image, and fal.ai — including background removal and composition.

Video Generation

Produces videos via Sora (OpenAI), Veo (Google), and Seedance 1.5 Pro (fal.ai). Also edits and combines existing clips.

Reference Finder

Audio analysis and reference track discovery using vector search (ChromaDB), sounddevice, and loudness normalization.

Ableton Dream Agent

Nightly song sketch generation, Ableton Live Extension code, and feedback loop tailored for house music production.

​
Real examples

Paste any of these directly into your terminal after running openswarm:
  • “Create a complete investor pitch for OpenSwarm” → Full deck + executive summary + market research
  • “Research my top 5 competitors and write 3 SEO-optimized blog posts” → Competitive analysis + keyword research + publish-ready content
  • “Analyze this data and create a quarterly report with charts” → Data insights + visualizations + formatted document
  • “Generate a product launch video with animations” → Professional video with graphics and transitions
  • “Build me a marketing campaign for Q2” → Strategy doc + creative assets + implementation timeline
Each prompt kicks off the Orchestrator, which breaks the work into parallel tracks and hands them off to the relevant specialists simultaneously. The results are assembled into a single cohesive output.

​
Multi-provider support

OpenSwarm supports three AI providers out of the box. Set the DEFAULT_MODEL environment variable to switch between them — all agents read it at startup via config.py.
ProviderDefault modelRouting
OpenAIgpt-5.2Native (no slash in model ID)
Anthropiclitellm/claude-sonnet-4-6Routed through LiteLLM
Google Geminilitellm/gemini/gemini-3-flashRouted through LiteLLM
Any model string containing a / is automatically routed through LiteLLM via agency_swarm.LitellmModel. OpenAI model IDs contain no slash and pass through unchanged. The setup wizard sets DEFAULT_MODEL for you when you run openswarm for the first time. Image and video generation agents use provider-specific models regardless of DEFAULT_MODEL — Gemini image models require GOOGLE_API_KEY, Sora requires OPENAI_API_KEY, and Seedance requires FAL_KEY.

Build docs developers (and LLMs) love

Get started for freeTalk to us