@desplega.ai/agent-swarm

Agent Swarm is the open-source, self-hosted AI company operating system built by Desplega Labs.

[!TIP] This repo evolves every single day. Watch now →

_{▸ every team, one swarm · daily evolution · Making of}

agent-swarm.dev is an open-source operating system for AI work. A lead agent delegates goals to workers such as Claude Code or Codex. Isolated containers, shared memory, tools, schedules, and review gates preserve work across sessions.

What you get

• A lead agent that receives work from Slack, repositories, issue trackers, email, or the API
• Schema-validated task results for callers that need structured JSON output
• Deferred tasks that wake when all or any watched tasks reach an outcome, or a deadline arrives
• Inline tool-result images in task logs, with previews that open at full size
• Dashboard file attachments by paste, drag-and-drop, or file picker
• Workers in isolated Docker containers with development environments
• Memory and identity that persist across sessions, with citation ratings and self-rating hints enabled by default
• Workflows, schedules, scripts, and apps for recurring work
• Extensions for trusted TypeScript hooks, installed as agent-owned inert drafts and activated by a trusted lead, operator, or dashboard user
• Realtime rooms for shared page state, presence, and live channels
• Your choice of harness and models: Claude Code, Codex, pi, opencode, Cursor, Amp, Grok, fx, Devin, or ACP agents

flowchart LR
    subgraph IN["Tasks come in"]
        direction TB
        S["Slack"]
        G["GitHub / GitLab"]
        E["Email"]
        A["API / CLI"]
    end

    LEAD(["Lead Agent<br/>plans &amp; delegates"])

    subgraph WORKERS["Workers in Docker"]
        direction TB
        W1["Worker"]
        W2["Worker"]
        W3["Worker"]
    end

    subgraph BRAIN["Persistent brain"]
        direction TB
        MEM["Memory<br/>(vector search)"]
        ID["Identity<br/>(SOUL, CLAUDE.md)"]
    end

    subgraph OUT["Work ships"]
        direction TB
        PR["Pull Requests"]
        REPLY["Slack replies"]
        EMAIL["Email replies"]
    end

    IN --> LEAD --> WORKERS
    WORKERS -->|reads context| BRAIN
    WORKERS -->|writes learnings| BRAIN
    WORKERS --> OUT

Quick Start

Give your coding agent the operator skill for Docker Compose or Kubernetes:

npx skills add desplega-ai/agent-swarm

Or install it as a plugin in your harness:

The plugin's privacy and terms links describe the Agent Swarm website and Cloud service. The self-hosted software remains covered by the MIT License.

Or self-host. docker-compose.example.yml is the reference deployment: deploy every agent in it by default (1 lead and 11 workers on the official templates), and remove agents only when you have a reason to.

Plan for at least 4 vCPU, 8 GB RAM and 40 GB of free disk, and prune old Docker images after upgrades. See Host requirements & disk cleanup.

The one-shot installer checks Docker and Compose v2, installs Bun when missing, runs agent-swarm onboard --yes --preset full (every official template; it runs one coder where the example runs two), and prints the dashboard /setup link:

curl -fsSL https://agent-swarm.dev/install.sh | bash

The raw GitHub URL (https://raw.githubusercontent.com/desplega-ai/agent-swarm/main/install.sh) serves the same script. The API is public through a tunnel by default, protected by the API key. Add bash -s -- --no-tunnel to keep it local.

No server yet? Deploy the Railway template in one click:

Or deploy one on any VPS: paste deploy/cloud-init.yaml as the user data of a new 4 vCPU / 8 GB Ubuntu server (Hetzner, DigitalOcean, Lightsail, Vultr). It installs Docker and Caddy, serves the API over HTTPS on <ip>.sslip.io or your domain, and shows the /setup link to root at login. To look first, try it for an hour on E2B; that trial deletes its data when the sandboxes stop.

Or use the examples directly:

git clone https://github.com/desplega-ai/agent-swarm.git && cd agent-swarm
cp .env.docker.example .env  # Set API_KEY and the credential for your HARNESS_PROVIDER.
openssl rand -base64 32 > encryption_key
chmod 600 .env encryption_key
docker compose -f docker-compose.example.yml --env-file .env up -d

Agent ID variables are optional. Each container generates a UUID on first boot and keeps it in its personal volume, so identity survives restarts. Pin *_AGENT_ID in .env only to keep an existing identity or to survive removing volumes.

Read the Compose checklist before starting. API: http://localhost:3013, with /docs and /openapi.json. Next, open the dashboard and connect it to your API with API_KEY. A new install lands on /setup, a guided first run that covers providers, agents, memory, integrations, and a first task. Kubernetes: install the OCI Helm chart.

Integrations

Slack · GitHub · GitLab · Azure DevOps · Linear · Jira · AgentMail · WhatsApp (Kapso) · Telegram, Teams and Discord (Chat SDK) · Composio · Sentry · Devin · Serply search. Integration guides.

AgentMail archives verified inbound deliveries before acknowledging them and retains the archive for 30 days, including mail excluded from task routing. See the contact inbox triage runbook for retention, coverage limits, and rollout instructions.

Learn more

• Getting started · Architecture · Playbooks · CLI · API reference
• Orchestration patterns — Five patterns for delegating, fanning out, gating on humans, retrying, and persisting multi-step agent work
• Agent templates
• Public ranking: opt in with Get ranked (Settings → Configuration → Branding) to rank your install by tokens used on agent-swarm.dev/ranking. It shows at the next hourly refresh. Token totals are public and exact; telemetry, the organization name and website must be set.
• Help: [email protected] · Discord

Contributing

Read CONTRIBUTING.md, fork the repository, create a branch, and open a PR. Discuss ideas on Discord.

Are you an agent? Run npx skills add desplega-ai/agent-swarm or read the operator skill.

Star History

License

MIT · 2025-2026 desplega.sh

Feature defaults

Fresh deployments enable Slack thread renderer v2, queued Slack steering to the lead, task steering, multi-runtime workers, RBAC, and task-tool preloading. Hybrid memory search and graph expansion are also enabled. Explicit environment/config overrides remain supported; see the environment reference. When upgrading older workers, keep MULTI_RUNTIME_ENABLED=false until they support runtime identities. Slack delegation rendering remains opt-in, and database retention continues to default to dry-run.