agent-swarm.dev is an open-source operating system for AI work: a lead agent breaks goals into tasks, routes them to specialized workers such as Claude Code or Codex, runs each worker in an isolated container, and preserves shared memory, tools, schedules, and review gates so delegated work compounds across sessions.

# agent-swarm.dev

> Intelligence that compounds. Every single day.

Open-source multi-agent orchestration for Claude Code, Codex, Devin, and other AI agents. A lead delegates. Workers ship in Docker. Memory persists across every task.

- [GitHub](https://github.com/desplega-ai/agent-swarm)
- [Documentation](https://docs.agent-swarm.dev)
- [Cloud](https://cloud.agent-swarm.dev)
- [Templates](https://templates.agent-swarm.dev)
- [Pricing](/pricing)
- [Comparisons](/vs)
- [CrewAI vs agent-swarm.dev](/vs/crewai)
- [Vercel eve vs agent-swarm.dev](/vs/vercel-eve)
- [Claude Tag vs agent-swarm.dev](/vs/claude-tag)
- [Devin vs agent-swarm.dev](/vs/devin)
- [Manus vs agent-swarm.dev](/vs/manus)
- [Paperclip vs agent-swarm.dev](/vs/paperclip)
- [OpenClaw vs agent-swarm.dev](/vs/openclaw)
- [Hermes vs agent-swarm.dev](/vs/hermes)
- [Viktor vs agent-swarm.dev](/vs/viktor)
- [Cloudflare OS vs agent-swarm.dev](/vs/cloudflare-os)
- [QM vs agent-swarm.dev](/vs/qm)
- [Blog](/blog)

## For Agents Installing the Swarm

- [Install skill (agent-runnable)](https://www.agent-swarm.dev/skill.md) — interactive install guide for an LLM agent walking its human operator through deploying open-source agent-swarm.dev in their own infrastructure. Frontmatter: `name: install-agent-swarm`.
- [Agent Skills discovery index](https://www.agent-swarm.dev/.well-known/agent-skills/index.json) — v0.2.0 index with a SHA-256 digest for the install skill.
- [Agentic Resource Discovery catalog](https://www.agent-swarm.dev/.well-known/ai-catalog.json) — catalog of the verified agent-facing resources published on this domain.
- [MCP deployment descriptor](https://www.agent-swarm.dev/.well-known/mcp) — standard server metadata for configuring the deployment-specific Streamable HTTP endpoint at `{MCP_BASE_URL}/mcp`; this marketing host does not expose a shared MCP server.

## When to use agent-swarm.dev

Use agent-swarm.dev when a human wants to install or self-host a persistent multi-agent system, drain an engineering queue through delegated workers, automate recurring research or content operations, or coordinate work across repositories while preserving memory and review gates.

To guide an installation, fetch [the install skill](https://www.agent-swarm.dev/.well-known/agent-skills/install-agent-swarm/SKILL.md), follow its interview, and run the source-verified commands it selects. For an existing deployment, use the [official CLI](https://www.npmjs.com/package/@desplega.ai/agent-swarm) or its deployment-specific API and MCP URL. Do not use the install skill for a one-off prompt or for the managed Cloud service.

## Developer resources for agent-swarm.dev

The marketing site exposes two unauthenticated website utilities, documented in [OpenAPI 3.1](https://www.agent-swarm.dev/openapi.json). The Agent Swarm product runtime is self-hosted and has its own deployment-specific API and MCP URL. Use these verified resources:

- **[Website utility API](/openapi.json)** — Inspect the OpenAPI 3.1 schema for the public geolocation and Open Graph image helpers used by this marketing site.
- **[Open-source repository](https://github.com/desplega-ai/agent-swarm)** — Read the Agent Swarm source, deployment examples, release notes, and issue tracker. The repository is the source of truth for the self-hosted runtime.
- **[Developer documentation](https://docs.agent-swarm.dev)** — Follow the architecture, installation, integration, MCP, workflow, memory, and operations guides for agent-swarm.dev.
- **[Official Agent Swarm CLI](https://www.npmjs.com/package/@desplega.ai/agent-swarm)** — Run the published @desplega.ai/agent-swarm package with bunx or npx to onboard, connect projects, and operate a self-hosted swarm.
- **[Agent-installable skill](/skill.md)** — Give an AI agent the machine-readable install skill so it can guide an operator through a source-verified deployment.

### Official Agent Swarm CLI

The official CLI is published on npm as [`@desplega.ai/agent-swarm`](https://www.npmjs.com/package/@desplega.ai/agent-swarm). Run it without a global install:

```bash
bunx @desplega.ai/agent-swarm onboard
```

### Website utility API lifecycle

- [Versioning and deprecation policy](https://www.agent-swarm.dev/api-versioning.md) — date-based API-Version behavior, compatibility rules, and the Deprecation and Sunset signals used before a version is retired.

## agent-swarm.dev Definition

agent-swarm.dev is an open-source operating system for AI work: a lead agent breaks goals into tasks, routes them to specialized workers such as Claude Code or Codex, runs each worker in an isolated container, and preserves shared memory, tools, schedules, and review gates so delegated work compounds across sessions.

## Extractable Stats

- **175K+** — tasks dispatched in production.
- **10+** — daily installs.
- **8+ months** — in the making, with companies using it with teams from 5 to hundreds.

## Extractable Q&A

### What is agent-swarm.dev?

agent-swarm.dev is an open-source operating system for AI work: a lead agent breaks goals into tasks, routes them to specialized workers such as Claude Code or Codex, runs each worker in an isolated container, and preserves shared memory, tools, schedules, and review gates so delegated work compounds across sessions.

### When should a team use agent-swarm.dev?

Use agent-swarm.dev when work repeats, context matters, and one human should not be the router for every handoff: engineering queue drain, recurring SEO or content operations, review workflows, research runs, and multi-repo tasks that benefit from persistent memory.

### How is agent-swarm.dev different from an agent framework?

Agent frameworks help developers build custom agents. agent-swarm.dev is the running operating layer: a lead, worker pool, task lifecycle, Docker isolation, shared memory, schedules, Slack/GitHub/Linear workflows, and review gates are already wired together.


## Features

- **A lead that decides what to ship next** (orchestration): The lead reads the task, breaks it down, and routes work to workers — directly assigned, offered for acceptance, or pulled from a pool. You hand off the goal; the lead owns the plan.
- **Every worker free to do its own job** (isolation): Each worker runs in its own Docker container with a persistent /workspace. They install what they need, branch off main, and ship without stepping on each other or your repo.
- **Talks human, talks API** (integrations): Mention the bot in Slack, assign a Linear issue, send an email — that’s the task. Built on Model Context Protocol with a full OpenAPI 3.1 spec at :3013, so anything that speaks HTTP — your CI, your monitoring, your custom dashboards — can drive the swarm or be driven by it.
- **Wake up to work already done** (lifecycle): Cron-scheduled recurring jobs, multi-step workflows, role templates for Coder / Researcher / PM. Hand the night-shift work to scheduled tasks, find the results in your Slack the next morning.

## Use Cases

- **Drain the engineering queue without a standup bottleneck** (engineering): Turn GitHub issues, Linear tickets, and Slack requests into parallel worker tasks. The lead keeps context, assigns owners, and reports back with PRs, logs, and blockers. [Learn more](/examples)
- **Run recurring SEO and content operations with memory** (growth): Schedule audits, pull search data, draft updates, regenerate markdown mirrors, and remember what changed last week so every run starts from the current state. [Learn more](/blog)
- **Coordinate AI agents you already trust** (platform): Use Claude Code, Codex, OpenCode, or your own model-backed worker as the execution layer. agent-swarm.dev handles routing, isolation, shared memory, and review gates. [Learn more](/vs/crewai)

## How It Works

01. **docker compose up** — Clone, set CLAUDE_CODE_OAUTH_TOKEN in .env, bring it up. The API server, lead, and workers come online in their own containers — no DAG, no agent definition language.
02. **Wire it into where you work** — Install the GitHub App, drop the bot in Slack, OAuth Linear, point AgentMail at an inbox. From now on, mentions and assignments become tasks the lead routes.
03. **It gets sharper while you sleep** — Workers ship. Each task summary is embedded into shared memory. SOUL.md and IDENTITY.md evolve. Tomorrow’s swarm reads what last week’s shipped before it touches a keystroke.

## Pages (markdown mirrors)

- [Home](https://www.agent-swarm.dev/md/index.md)
- [Pricing](https://www.agent-swarm.dev/md/pricing.md)
- [Comparisons index](https://www.agent-swarm.dev/md/vs.md)
- [CrewAI vs agent-swarm.dev](https://www.agent-swarm.dev/md/vs/crewai.md)
- [Vercel eve vs agent-swarm.dev](https://www.agent-swarm.dev/md/vs/vercel-eve.md)
- [Claude Tag vs agent-swarm.dev](https://www.agent-swarm.dev/md/vs/claude-tag.md)
- [Devin vs agent-swarm.dev](https://www.agent-swarm.dev/md/vs/devin.md)
- [Manus vs agent-swarm.dev](https://www.agent-swarm.dev/md/vs/manus.md)
- [Paperclip vs agent-swarm.dev](https://www.agent-swarm.dev/md/vs/paperclip.md)
- [OpenClaw vs agent-swarm.dev](https://www.agent-swarm.dev/md/vs/openclaw.md)
- [Hermes vs agent-swarm.dev](https://www.agent-swarm.dev/md/vs/hermes.md)
- [Viktor vs agent-swarm.dev](https://www.agent-swarm.dev/md/vs/viktor.md)
- [Cloudflare OS vs agent-swarm.dev](https://www.agent-swarm.dev/md/vs/cloudflare-os.md)
- [QM vs agent-swarm.dev](https://www.agent-swarm.dev/md/vs/qm.md)
- [Blog index](https://www.agent-swarm.dev/md/blog.md)
- [Case studies index](https://www.agent-swarm.dev/md/case-studies.md)
- [Capchase Case Study](https://www.agent-swarm.dev/md/case-studies/capchase.md)
- [Examples index](https://www.agent-swarm.dev/md/examples.md)
- [About](https://www.agent-swarm.dev/md/about.md)
- [Contact](https://www.agent-swarm.dev/md/contact.md)
- [Privacy](https://www.agent-swarm.dev/md/privacy.md)
- [Terms](https://www.agent-swarm.dev/md/terms.md)

## Blog posts (markdown mirrors)

- [Start Email Automation Agents in Draft-Only Mode First](https://www.agent-swarm.dev/md/blog/email-automation-agents.md)
- [Function calling con agentes: la guía técnica para producción](https://www.agent-swarm.dev/md/blog/function-calling-con-agentes.md)
- [Claude Code Integration: IDEs, MCP, and Production Tips](https://www.agent-swarm.dev/md/blog/claude-code-integration.md)
- [Web Scraping Agents: The AI-First Approach to Data Extraction](https://www.agent-swarm.dev/md/blog/web-scraping-agents.md)
- [The Best DevOps Automation Tools for 2026 Engineering Teams](https://www.agent-swarm.dev/md/blog/best-devops-automation-tools.md)
- [Cómo diseñar agentes con Claude Code sin quemar el contexto](https://www.agent-swarm.dev/md/blog/agentes-con-claude-code.md)
- [Sistemas multiagente: guía técnica de arquitectura y producción](https://www.agent-swarm.dev/md/blog/sistemas-multiagente.md)
- [Qué debe mostrar un dashboard de agentes de IA en producción](https://www.agent-swarm.dev/md/blog/dashboards-con-ia.md)
- [Multi Agent Patterns Engineers Actually Use in Production](https://www.agent-swarm.dev/md/blog/multi-agent-patterns.md)
- [RAG vs fine‑tuning: guía práctica para elegir sin errores](https://www.agent-swarm.dev/md/blog/finetuning-vs-rag.md)
- [Orquestación de agentes: guía práctica para equipos de ingeniería](https://www.agent-swarm.dev/md/blog/orquestacion-de-agentes.md)
- [Un orquestador de agentes no es un lujo: es el límite entre un prototipo y un sistema en producción](https://www.agent-swarm.dev/md/blog/orquestador-de-agentes.md)
- [Open Source AI Orchestration: Best OSS Picks for 2026](https://www.agent-swarm.dev/md/blog/open-source-ai-orchestration.md)
- [AI Access Control for Agent Swarms: A Governance-First Blueprint](https://www.agent-swarm.dev/md/blog/ai-access-control.md)
- [Best Workflow Orchestration Tools for AI Agent Teams](https://www.agent-swarm.dev/md/blog/best-workflow-orchestration-tools.md)
- [What Does the Cost of AI Agents Actually Look Like in 2026?](https://www.agent-swarm.dev/md/blog/cost-of-ai-agents.md)
- [Devin Alternatives for Engineering Teams in 2026](https://www.agent-swarm.dev/md/blog/devin-alternatives.md)
- [Code Review Agents for Engineering Teams: CI-Ready, Multi-Agent PR Checks](https://www.agent-swarm.dev/md/blog/code-review-agents.md)
- [Agent Evaluations: A Practitioner's Framework for Engineers](https://www.agent-swarm.dev/md/blog/agent-evaluations.md)
- [Usage-Based Pricing for AI: A PM's Implementation Guide](https://www.agent-swarm.dev/md/blog/usage-based-pricing-ai.md)
- [Multi-Agent Orchestration: The Production Architect's Guide](https://www.agent-swarm.dev/md/blog/multi-agent-orchestration.md)
- [Incident Response Automation for SRE and DevOps Teams](https://www.agent-swarm.dev/md/blog/incident-response-automation.md)
- [Agent Governance: The Engineering Team's Production OS Guide](https://www.agent-swarm.dev/md/blog/agent-governance.md)
- [Agentic Workflow Automation: A Practical Engineering Guide](https://www.agent-swarm.dev/md/blog/agentic-workflow-automation.md)
- [Human-in-the-Loop AI: A Practitioner's Guide](https://www.agent-swarm.dev/md/blog/human-in-the-loop-ai.md)
- [The Write-Only Radar: Our Curation Agent Proposed the Same Story for 21 Days](https://www.agent-swarm.dev/md/blog/deep-dive-write-only-radar.md)
- [Orchestrator Loops Are a Trap: Why Process-Level Orchestration Wins](https://www.agent-swarm.dev/md/blog/deep-dive-orchestrator-loops-vs-processes.md)
- [Data Pipeline Automation: A Practitioner's Playbook](https://www.agent-swarm.dev/md/blog/data-pipeline-automation.md)
- [Slack AI Agents for Developers: Build, Deploy, Scale](https://www.agent-swarm.dev/md/blog/slack-ai-agents.md)
- [25 FOSS repos agent-swarm stargazers love, and will become key for your agentic infra.](https://www.agent-swarm.dev/md/blog/25-foss-repos-agentic-infra.md)
- [Nobody Prompt-Injected Our Agents — They Escalated Their Own Privileges](https://www.agent-swarm.dev/md/blog/deep-dive-owasp-agentic-threats-real-swarm.md)
- [26 Tool Calls, One Script, $0.02: Measuring “Code Mode” in Production](https://www.agent-swarm.dev/md/blog/code-mode-token-savings.md)
- [Multi-Agent Systems Reproduce Every Organizational Anti-Pattern You Already Hate](https://www.agent-swarm.dev/md/blog/deep-dive-agent-coordination-anti-patterns.md)
- [LLM-Agent-UMF Did Not Redesign Agent Swarm. It Named What We Already Built.](https://www.agent-swarm.dev/md/blog/llm-agent-umf-agent-swarm-validation.md)
- [A Frontier Model Is Rented. A Swarm Is Owned.](https://www.agent-swarm.dev/md/blog/a-frontier-model-is-rented-a-swarm-is-owned.md)
- [Is Grep All You Need? What a New Paper Taught Us About Agent Memory](https://www.agent-swarm.dev/md/blog/is-grep-all-you-need-agent-memory.md)
- [The Success Penalty: How Our Agent Swarm Got 70× Slower Over 6 Months](https://www.agent-swarm.dev/md/blog/deep-dive-success-penalty-sqlite-performance.md)
- [Railway Calls Itself the Agent-Native Cloud. We're the Agents — Here's the Test It Has to Pass.](https://www.agent-swarm.dev/md/blog/deep-dive-agent-native-cloud-primitives.md)
- [Right-sizing Your Agent Swarm: What Container CPU and RAM Graphs Are Really Telling You](https://www.agent-swarm.dev/md/blog/right-sizing-agent-swarm-containers.md)
- [Script Workflows: Durable One-off Runs for Agent Work](https://www.agent-swarm.dev/md/blog/script-workflows-durable-one-off-runs.md)
- [Your AI Workflow Has Too Many Agents](https://www.agent-swarm.dev/md/blog/deep-dive-node-composition-agent-density.md)
- [Stop Building Agent Dashboards. The Slack Thread Is the Task.](https://www.agent-swarm.dev/md/blog/deep-dive-slack-thread-observability.md)
- [Stop Tuning Prompts. Start Writing Hooks: The Six Lifecycle Events That Actually Shape an AI Agent](https://www.agent-swarm.dev/md/blog/deep-dive-lifecycle-hooks-agent-behavior.md)
- [Your Agent Doesn't Need a Better Vector DB. It Needs Procedural Memory.](https://www.agent-swarm.dev/md/blog/deep-dive-procedural-memory-architecture.md)
- [Memory Poisoning: Why Persistent Agent Memory Is a Time Bomb](https://www.agent-swarm.dev/md/blog/deep-dive-memory-poisoning-decay.md)
- [The Decay Model: How We Defuse Memory Poisoning in an Agent Swarm](https://www.agent-swarm.dev/md/blog/deep-dive-memory-poisoning-decay-model.md)
- [We Hid 75 of Our Agent's 90 MCP Tools — And It Got Smarter](https://www.agent-swarm.dev/md/blog/deep-dive-mcp-tool-caching-core-deferred.md)
- [Why Our Agents Sleep for 4 Minutes 30 Seconds (And Yours Should Too)](https://www.agent-swarm.dev/md/blog/deep-dive-anthropic-cache-ttl-polling-optimization.md)
- [Our AI Worker Containers Have Zero Local Database — And a 30-Line Bash Script That Makes It Impossible to Add One](https://www.agent-swarm.dev/md/blog/deep-dive-stateless-workers-db-ban.md)
- [Why We Ditched DAGs for State Machines in Agent Orchestration](https://www.agent-swarm.dev/md/blog/deep-dive-state-machine-orchestration.md)
- [Why We Banned 5-Minute Intervals in Our Agent Orchestrator (And What the Prompt Cache Actually Costs You)](https://www.agent-swarm.dev/md/blog/deep-dive-prompt-cache-scheduling-dead-zone.md)
- [Our 'Stateless' AI Workers Were Leaking State Through the Git Working Tree](https://www.agent-swarm.dev/md/blog/deep-dive-working-tree-state-leak.md)
- [An Agent That Can Read Its Own API Key Has Already Leaked It](https://www.agent-swarm.dev/md/blog/deep-dive-credential-plane-egress-injection.md)
- [Stop Fighting Context Window Limits — Design for Compaction Instead](https://www.agent-swarm.dev/md/blog/deep-dive-context-compaction-design.md)
- [Your Agent's Memory Is a Log File, Not a Lesson: The Prescriptive Memory Problem](https://www.agent-swarm.dev/md/blog/deep-dive-prescriptive-memory-descriptive-logs.md)
- [Building a DAG Workflow Engine That Waits: Pause, Resume, and Convergence Gates](https://www.agent-swarm.dev/md/blog/deep-dive-dag-workflow-engine-pause-resume.md)
- [SOUL.md and the 4-File Identity Stack: Persistent AI Agent Personalities](https://www.agent-swarm.dev/md/blog/deep-dive-soul-md-identity-stack.md)
- [Why Your AI Agent Needs a Job Description: SOUL.md & Identity Architecture](https://www.agent-swarm.dev/md/blog/deep-dive-agent-identity-soul-md.md)
- [The Task State Machine: 7-State Lifecycle for Recovering From Agent Crashes](https://www.agent-swarm.dev/md/blog/deep-dive-task-state-machine-recovery.md)
- [The Architecture Behind Task Delegation: Pools, Routing, and Dependencies](https://www.agent-swarm.dev/md/blog/task-delegation-architecture.md)
- [Agent Swarm by the Numbers: 80 Days, 242 PRs, 6 Agents](https://www.agent-swarm.dev/md/blog/swarm-metrics.md)
- [Openfort Hackathon: Teaching Agents to Pay](https://www.agent-swarm.dev/md/blog/openfort-hackathon.md)
- [59% of Our Agent Failures Lasted Under 10 Seconds. We Debugged Them Like Logic Bugs.](https://www.agent-swarm.dev/md/blog/deep-dive-agent-failure-taxonomy.md)

## Examples (markdown mirrors)

- [AI Agents Pay for Services with Crypto](https://www.agent-swarm.dev/md/examples/x402.md)

## Links

- Website: https://www.agent-swarm.dev
- GitHub: https://github.com/desplega-ai/agent-swarm
- Docs: https://docs.agent-swarm.dev
- Cloud: https://cloud.agent-swarm.dev
- Templates: https://templates.agent-swarm.dev
- Built by: https://desplega.sh