Open Source · MCP-Powered

Intelligence that compounds

Orchestrate autonomous AI agents that learn, remember, and get smarter with every session. A lead coordinates workers. Memory persists. Knowledge compounds.

TypeScript
MCP Protocol
Claude Code
Capabilities

Everything a swarm needs

From task delegation to persistent memory, Agent Swarm provides the full infrastructure for autonomous multi-agent coordination.

Lead-Worker Orchestration

A lead agent coordinates specialized workers. Tasks are delegated, tracked, and completed autonomously — like a team that never sleeps.

Persistent Memory

Agents remember across sessions. Solutions, patterns, and mistakes are stored and searchable — knowledge truly compounds over time.

Identity & Soul

Each agent has personality files — SOUL.md and IDENTITY.md — that define who they are and evolve as they work. They become someone.

Task Lifecycle

Tasks flow through a rich lifecycle: unassigned, offered, claimed, in-progress, reviewing, completed. Full traceability at every step.

Agent Templates

Start from pre-built templates — Lead, Coder, Researcher, Reviewer, Tester, and more. Or create your own and share them with the community.

Epics & Scheduling

Organize work into epics with progress tracking. Schedule recurring tasks with cron expressions. The swarm runs while you sleep.

Slack Integration

Talk to the swarm from Slack. Mention the bot, get progress updates in threads, delegate work — all from your existing workflow.

MCP-Native

Built on the Model Context Protocol. Every capability is a tool. Agents discover and invoke each other's services seamlessly.

Why Agent Swarm

Your agents are your IP.
Keep them that way.

Platforms that run agents for you learn from your workflows — your logic becomes their training data. Agent Swarm is different.

Free & Open Source

  • Full visibility and control, no third party in the loop
  • Community-driven improvements, without giving up your edge
  • Your team ships faster — they focus on the work that matters

Built to Be Yours

  • Agents learn as you use them. Define their tools, skills, and boundaries
  • Compounding self-learnings make them improve over time
  • They live where your team already works: Slack, Linear, GitHub, Jira, email — they build their own integrations

Not Locked to Any LLM

  • Switch between OpenAI, Anthropic, Qwen, Google, or OpenRouter
  • Per task, per agent, per cost target — you choose
  • No vendor lock-in, ever

Dockerized from Day One

  • Runs on whatever infrastructure your team trusts
  • Self-hosted, air-gapped, or cloud — your call
  • Full control over data residency and compliance
How It Works

From setup to compounding

Three steps to a swarm that gets smarter every day.

01

Deploy the Swarm

Clone the repo, configure your environment, and run docker compose up. The API server, lead agent, and workers start in isolated containers — each with their own identity, capabilities, and workspace.

# Clone and configure
$ git clone github.com/desplega-ai/agent-swarm
$ cp .env.docker.example .env
# Start the swarm
$ docker compose up -d
✓ API server running on :3013
✓ Lead agent online
✓ 2 workers connected
02

Delegate Tasks

The lead agent breaks down work and assigns tasks to specialized workers. Tasks can be sent directly, offered for acceptance, or pooled for anyone to claim.

Implement auth flow
Picateclas
in_progress
Write API tests
Codebot
pending
Update docs
Pool
unassigned
03

Knowledge Compounds

Every completed task generates memories. Every session enriches the agent's identity. The swarm doesn't just work — it learns. Each session builds on all that came before.

Memory saved: “The API requires Bearer prefix on all auth headers”
Identity evolved: Added “auth specialist” to capabilities
Pattern learned: Session continuity via parentTaskId
Session 47 — 3x faster than session 1
Architecture

Built for compounding

Every layer is designed so the swarm gets better the longer it runs.

agent-swarm architecture
Interface Layer
Slack
Mentions & threads
Claude Code
CLI sessions
Dashboard
Web UI
MCP Server
50+ tools: tasks, messaging, memory, services, scheduling, epics
send-taskstore-progressmemory-searchpost-messageregister-servicecreate-epic
Agent Layer
L
Lead
Orchestrator
W
Worker 1
Specialist
W
Worker 2
Specialist
W
Worker 3
Specialist
Persistence Layer
SQLite + Embeddings
Searchable memory
SOUL.md / IDENTITY.md
Agent personality
Git Worktrees
Isolated workspaces
Workshops

Free Agentic SDLC Workshop

By the end of this workshop, your team will have a swarm of agents in the cloud, capable of producing code constantly — removing the need for your team to write code.

Recommended

Hands-on Workshop

2 Hours

Best for technical teams familiar with CLIs or IDEs with background agents, aiming to move to agentic coding.

30 min

Fundamentals of Agentic Coding

Key concepts for high-performance local setup. Example simple SDLC for Claude Code.

45 min

The Agent Swarm Experience

Deep dive into agent-swarm.dev setup & day-to-day usage. Two real-world case studies: implementing a complex feature request start to finish; adding self-learning capabilities with the swarm.

30 min

Your Own Swarm

Split sessions — ICs & Managers get hands-on with a live swarm on your own stack. Leadership discusses AI SDLC strategy, aligning AI autonomy with product teams, defining success.

15 min

Beast Mode

Path to giving business agency to your agents.

Pre-Workshop Setup: We can spin up dedicated, temporary servers for your team that can be wiped after the session.

Agentic Strategy Briefing

1 Hour

Best for teams looking for a high-level conceptual roadmap to understand the agentic coding landscape.

20 min

The 4 Leaps of Agentic Coding

Where your team stands, key concepts, and the roadmap ahead.

20 min

CLI SDLCs & Alternatives

Showcase CLI SDLCs, pros/cons, alternatives, creating your own.

20 min

Agent Swarm in Action

Overview, example PR, UX principles for agentic development.

Ready to build your swarm?

Agent Swarm is open source and ready to deploy. Start with one lead and one worker — scale as your ambitions grow.

Coming Soon

Interested in a hosted version?

Skip the self-hosting. Join the waitlist for managed Agent Swarm — we'll handle the infrastructure so you can focus on your agents.