Contact us

Giant Swarm Agent Platform

Run AI Agents in Production. On Your Terms.

The Giant Swarm Agent Platform is an open, sovereign platform for building and running enterprise AI agents on any Kubernetes cluster: cloud, on-prem, edge, or air-gapped.

Talk to us Watch the demo
  • 100% open
    source
  • Every tool
    call audited
  • Built in
    the EU

Enterprises have trusted us with their critical infrastructure for over a decade. 12+ years in production Kubernetes. Co-creators of the CNCF AI Conformance program.

One platform. Two products. Start with either.

Two products that work alone or together. Start with the one that solves today's problem.

How Tool Access works Your agents and MCP clients reach everything through one MCP Gateway, which handles single sign-on through your identity provider, scoped access per user, and a full audit trail. Behind the gateway sit your MCP servers: your own tools and systems, and the MCP Workflow Engine, which is an MCP server like any other. Workflows are made by a loop: an agent runs a task, the outcome is captured, the repeated pattern becomes a workflow. Those workflows then become MCP tools the gateway can call, which makes every run cheaper. Your agents and MCP clients MCP Gateway Single sign-on | Scoped access, per user | Full audit trail MCP servers Your tools and systems MCP Workflow Engine How a workflow gets made New workflows become MCP tools. Agent runs a task Outcome captured Pattern becomes a workflow cheaper every run

Tool Access: one governed path to every tool

Your teams already connect agents to internal systems through MCP servers: each one its own access point, its own credentials, its own blind spot. Tool Access puts them all behind a single governed gateway, and loads tools lazily, so your agents' context stays lean however many you connect.

Repeated work becomes a deterministic workflow: code replaces the LLM on known paths, so the same task costs less every time it runs.

Read the docs View on GitHub

Built on open source, developed in the open

  • kagent
  • agentgateway
  • muster

See the Agent Platform in action

A recorded run against a genuinely broken checkout service. We build the agent in the platform UI - a job description, an approved model, skills pulled from the team's shared repo - deploy it onto Kubernetes as a Helm chart, and watch it work out that the app is being OOMKilled. No staged demo, no script on someone's laptop.

A recorded run against a genuinely broken checkout service.

Giant Swarm Agent Platform Product walkthrough · 3 min

Compose

Compose an agent from shared building blocks, in minutes.

Built for the gap between your AI-native teams and everyone else

A few of your teams are already deep into agents. Their wins stay locked inside the team, while everyone else waits or improvises without governance. The Agent Platform turns one team's breakthroughs into everyone's starting point.

  • Cheaper every run

    Costs fall as usage grows, instead of scaling with it. The work an agent repeats stops being paid for by the token.

  • First agent in minutes

    Domain teams ship their own agents without waiting on the platform team, and without writing custom code.

  • Share what works

    One team's agents, skills, and integrations become available to every team, instead of being rebuilt behind every door.

  • One gateway for every tool

    Answer which agent touched what, and on whose behalf, for any tool call, without chasing down per-server logs.

  • Everything reviewable

    Read what an agent is allowed to do before it ever runs, and roll it back like anything else you ship.

  • Runs anywhere

    Wherever your data has to stay, your agents run there too, including regulated and fully disconnected sites.

Governed by design

Agents on the platform run contained: every agent lives in its own container in a restricted Kubernetes environment, with a blast radius you declare and review before it ever runs.

What an agent can and cannot reach An agent sits inside a controlled blast radius formed by four controls: pod, identity, network and tools. Inside the boundary it can reach erp.read and invoices.write. Outside it, payroll is not in its allowlist, the production database has no egress route, and secrets are not permitted for its user - an attempt to reach them is refused at the boundary. Controlledblast radius Payroll not in allowlist Prod DB no egress route Secrets user not permitted erp.read invoices.write identity tools network pod Agent

  • toolsTools

    Allowlists at the gateway decide which tools an agent may call, per agent and per user.

  • networkNetwork

    Network policies control exactly where an agent can connect.

  • identityIdentity

    Agents act as the signed-in user, via on-behalf-of tokens, not shared credentials.

  • podPod

    Restricted security policies enforce isolation for every agent container, by default.

Single sign-on everywhere, through your existing identity provider. Audit trails and scoped access control are not add-ons; they are how the platform works.

Your infrastructure, your rules

The platform runs on any CNCF-conformant Kubernetes cluster, version 1.33 or newer, whether we manage it or you do. Your agents, your models, and your data stay inside your perimeter.

Everything is open source and standard Kubernetes resources, so there is no exit to negotiate: if we disappeared tomorrow, your agents would keep running.

No Kubernetes yet? One server is enough. We install, you start.

  • Cloud

    Any provider. We manage it, or you do.

  • On-prem

    Your own data centre, your own hardware.

  • Edge

    Factory floor or remote site.

  • Air-gapped

    No outside connection at all.

Your first use case live in production, in 4 weeks

Start as a founding customer and skip the empty platform. We deploy in your environment, pre-filled with curated agents, skills, and tool integrations, delivered as tested, signed images from our registry.

  1. Week 1

    Platform Setup & First Use Case

    We deploy in your environment and scope the first use case with you.

  2. Week 2

    Building Blocks & First Agents

    You and your FDE build the skills and MCP servers the use case needs.

  3. Week 3

    Agents in Production & Handover

    Your agents go live in production, and your team takes over from us.

  4. Week 4

    Tuning & Continuous Improvement

    We tune what runs, and leave your team able to keep improving it.

Founding customers get direct influence on the roadmap and a forward deployed engineer working alongside their team.

Talk to us

Read more

Frequently Asked Questions

What is the Giant Swarm Agent Platform?

The Giant Swarm Agent Platform is an open source, self-hosted platform for building and running enterprise AI agents on any Kubernetes cluster, with every tool call governed and audited. It consists of two products: Tool Access, a governed gateway to all your MCP tools, and Agent Runtime, where composable agents are built and run. Each works on its own, and they are stronger together.

How do you run AI agents on Kubernetes in production?

On the Giant Swarm Agent Platform, every agent runs as a declarative Kubernetes resource in its own contained environment, deployed and reviewed through GitOps like any other production workload. That means the practices your platform team already trusts (version control, review, rollback, policy enforcement) apply to agents from day one.

How can we give AI agents access to internal tools without creating new security risks?

The safest pattern is a single governed gateway between agents and tools, instead of point-to-point connections. On the Agent Platform, all MCP servers sit behind one gateway with single sign-on, per-user tool visibility, allowlists, and a complete audit trail, so an agent can only ever reach the tools its user is allowed to use. The gateway holds no shared service credentials: it acts on behalf of the signed-in user, through OAuth 2.0 on-behalf-of token exchange.

Does the Agent Platform work with our existing infrastructure?

Yes: the Giant Swarm Agent Platform runs on any CNCF-conformant Kubernetes cluster (version 1.33 or newer), in cloud, on-prem, edge, or air-gapped environments, and connects to your existing identity provider. You bring your own models, your existing MCP servers plug into the gateway, and the UI can integrate into an existing Backstage portal.

Show 5 more questions Show fewer questions
Do we get audit trails and scoped access control?

Yes: audit trails and scoped access control are core to the Agent Platform rather than add-ons. Every tool call is traced and attributed to the person the agent acted for, and access is scoped per user: each user, and each agent acting on their behalf, sees only the tools they are entitled to.

Can we run AI agents in regulated or air-gapped environments?

Yes: the platform is fully self-hosted and runs in air-gapped environments, on the edge, and on the factory floor. Your data, your models, and every agent interaction stay inside your perimeter, which is what makes agents viable for regulated industries and production systems in the first place.

The platform is open source. What do we pay for?

The full platform is open source with no feature gates; what you pay for is the head start and the ongoing curation. The open source install starts empty, while customers get the platform pre-filled with curated agents, skills, and tool integrations, refined on our own agent fleet and delivered as tested, signed images, with support and optional full management on top.

How is the Agent Platform different from hyperscaler agent offerings or building our own platform?

Hyperscaler and SaaS agent offerings are the fastest way to start, but your agents then depend on one vendor's cloud, jurisdiction, and pricing, and your data leaves your perimeter. Building your own platform keeps control, but it means evaluating and integrating a tool landscape that changes weekly, indefinitely. The Giant Swarm Agent Platform is the curated middle path: production-ready now, fully open source in your environment, with any model or tool swappable at any time.

Does the Agent Platform meet digital sovereignty and data-residency requirements?

The Giant Swarm Agent Platform is built for digital sovereignty: fully open source, self-hosted on your own infrastructure, working with models you host yourself, so your data never leaves your perimeter and no third-party cloud is required. It runs air-gapped where needed, and it comes from Giant Swarm, a European vendor.

Bring your agents into production, on founding-customer terms

A 30-minute call: live demo, and a first scoping of your use case.

Talk to us

Watch the demo View on GitHub