Build agents for your clients.
We run them.

Swirls is the platform for MSPs, agencies, and forward-deployed engineers. Declare agentic systems in .swirls files, deploy them per client, and govern every run.

Read the docs

The divide

Most agent projects never reach production.

The blocker is not the model. It is getting agents integrated with a client’s systems, governed against real policy, and running long after the pilot. That bridge is the work, and it is where most teams stall.

Swirls is built for that bridge. It is how managed service providers and forward-deployed engineers turn an experiment into a system a client can trust.

95%of organizations see zero return on their GenAI investment.MIT State of AI in Business 2025, via the Pax8 MIP Playbook.

Step 1

An agent in a dozen lines.

Teams of agents — a few lines, no boilerplate.

acme · concierge

Teams of agents — a few lines, no boilerplate.

agent.swirls

concierge › agent.swirls

secret provider_keys {
  label: "Provider keys"
  vars: [OPENROUTER_API_KEY]
}

agent concierge {
  label: "Concierge"
  secrets: provider_keys
  provider: openrouter
  model: "openai/gpt-4o-mini"
  // See tools.swirls for workflow definitions.
  tools: [search_orders]
  system: @ts {
    return "Help customers with orders, returns, and shipping updates."
  }
}

terminal

~/acme $swirls doctor

✓ 1 agent · 2 workflows · 2 triggers

Delivery model

Build it once. Run it for every client.

Swirls is the system of record for the agents you ship. Stand the work up once, then deploy it to each client, isolated and governed.

Build the system once

Declare agents, workflows, tools, and triggers in .swirls files. One codebase, version controlled, that captures how the work gets done.

Deploy it per client

Ship the same system to each client with a git push. A project per client, connections scoped per client, every tenant isolated.

Govern and expand

Watch every run, control what each agent can touch, and grow the engagement. The system compounds value instead of going stale.

Governance by design

Governed from the first engagement.

Embed compliance, security, and cost controls from day one. It is what separates a system a client can trust from a demo that never ships.

Scoped per run

Every connection is encrypted and scoped to a single run. One client's keys never reach another client's work.

Deny by default

Map a client org to roles, then grant or deny each agent and tool. Lock the system down when the work demands it.

Built in, not bolted on

Governance lives in the same files as the system. It ships and reviews with everything else, from the first engagement.

See how the security model works

Primitives

Use one or all.

Best-of-breed primitives, integrated as infrastructure. Assemble them into a system you run for every client.

Agents

Declarative LLM agents with tools, profiles, teams, and a persistent workspace — the agentic harness without the boilerplate.

agent concierge { … }

Workflows

Deterministic, typed DAGs of code, AI, and HTTP nodes. Durable execution with a schema at every edge.

workflow search_orders { … }

Connections

Brokered OAuth or bring your own keys. Encrypted and scoped to each run, so every client's systems stay isolated.

connection slack { … }

Access control

Map a client's org to roles, then grant or deny agents and tools. Deny by default when it matters.

policy { allow … }

Triggers

Forms, webhooks, and schedules bind to workflows. Anything can start a run.

webhook:inbound → run

Streams & views

Persist typed workflow output as records, then read it back or compose it into spreadsheet views.

stream scored_leads { … }

Disk

Mounted, persistent disks your agents and workflows can read, write, and run bash against.

disk workspace { … }

Channels

Bind an agent to Slack, Linear, Discord, or the web so it answers right in the conversation.

channel support_bot { … }

Customers

Built for the people building for everyone else.

Agents, workflows, triggers, connections. Compose them into a system you run for each client. BrewTech built an entire agency on Swirls.

Customer story

Swirls lets me build an entire agency on top of it. I take it and run.

Ben BrewerVP of Engineering, BrewTech IT

Read the BrewTech story

Industry: AI agent agency
Shape: Specialists under one orchestrator
Declared in: .swirls files

Ship your first client.

Author .swirls files. Deploy free to inspect your project in Swirls Cloud.

Browse the cookbook

Build agents for your clients.We run them.