desplega-ai/agent-swarm

Your Company Agentic Operating System

What it solves

Agent Swarm provides a system for companies to deploy a coordinated team of AI agents that handle complex business operations autonomously. It solves the problem of fragmented AI usage by creating a "compounding intelligence layer" where agents share memory and learnings, allowing them to perform roles like engineering, marketing, and customer support without starting from scratch every time.

How it works

The system uses a lead-worker architecture:

  1. Lead Agent: Acts as the coordinator that receives tasks from external channels (Slack, GitHub, Email, etc.), plans the execution, and delegates subtasks.
  2. Worker Agents: Execute the actual work within isolated Docker containers, providing full development environments (Node.js, Python, etc.).
  3. Persistent Brain: A shared memory system using vector search and identity files that allows agents to read context and write learnings back for future use.
  4. Integration Layer: Connects to tools like Jira, Linear, and Sentry to ingest tasks and ship results as Pull Requests or messages.

Who it’s for

It is designed for builders and companies wanting to move from "AI-first" to "AI-native" operations, specifically those needing automated feature development, lead prospecting, proactive customer support, or automated code health management.

Highlights

  • Isolated Execution: Workers run in Docker containers to ensure safe and consistent environments.
  • Compounding Memory: Hybrid vector and full-text recall allows agents to evolve their expertise and remember past sessions.
  • LLM-Agnostic: Supports various harnesses including Claude Code, OpenAI Codex, Devin, and raw LLMs.
  • Human-in-the-Loop: DAG-based workflow engine with approval gates for irreversible steps.
  • Multi-channel Input: Native integrations with Slack, GitHub, GitLab, WhatsApp, Email, Linear, and Jira.
  • Observability: OpenTelemetry traces and OTLP metrics for tracking costs and token usage.