Agent Platform for software teams

Your developers already use coding agents. Now your security team can say yes.

Coding agents run in a sandbox in your environment, with the network access, the models and the budget IT sets. Every tool call is logged and a person approves each rollout. See it with one of your repositories.

  1. SpecSpecification Assistant

    Someone from product describes a status page in one sentence. The assistant asks three questions and writes the spec.

  2. BuildCoding Agent

    Code and tests, written in a sandbox in your environment. Every command is in the session.

  3. ApproveDeployment Assistant

    Roll out status-page 1.4.0 to staging? It waits until a person allows it.

  4. Roll outRelease pipeline

    Version 1.4.0 rolls out region by region. The approval is in the audit trail.

Trusted by platform teams at

  • adidas
  • Börse Stuttgart
  • Deutsche Telekom
  • EGGER
  • Vodafone
  • WEPA

01 / Why

Coding agents already run on your developers' laptops.

Each developer picked a tool, pasted a key and pointed it at a repository. Nobody planned for security, cost or review.

Topic On a laptop today On the Agent Platform
Where the agent runs On the developer's machine, next to everything else on it. In one sandbox per session in your environment.
Keys A personal API key in a config file. A developer's own ChatGPT plan, company keys or a model you host, set per team.
Network access Any host the laptop can reach. The hosts IT lists. The platform blocks everything else.
What the agent ran Visible to one developer until the terminal closes. Every tool call, command and output stays in the session for the team and for audit.
Cost A separate bill per tool and per person. Per agent, per team and per model, with a budget for each team.
If you want to leave Each tool keeps its own history and settings. Open source and built in the EU. If the contract ends, the platform stays with you.

Already using Copilot or Cursor? They stay useful in the editor. The Agent Platform runs agents that work without a laptop, in your environment, under shared rules for access, cost and approval.

02 / How

From one sentence to staging.

The team that uses the three agents in the demo also built them. Each step shows its work and the rollout waits for a person. All data in the demo is fictional.

One feature4 steps

  1. Spec

    Specification Assistant

    Under 250 words

    Asks three questions, then writes goal, scope and acceptance criteria.

  2. Build

    Coding Agent

    Sandboxed

    Writes the code and the tests in a sandbox with the tools IT approved. Every file change and command shows up in the session.

  3. Approve

    Deployment Assistant

    Waiting for a person

    Summarizes the release and pauses until a person clicks Allow.

  4. Roll out

    Deployment Assistant

    Per region

    Rolls out region by region once a person approves the release.

03 / What

Each team picks the agent and the model. IT sets the rules once.

A developer connects the ChatGPT plan they already have, or the team uses company keys or a model you host. The agent stays the same.

The model is one card in More options.

For your security review

Every coding agent on the platform inherits these rules.

They apply whichever agent and model a team picks.

Coding agents

  • Codex
  • Claude Code

Models

  • A developer's own ChatGPT plan
  • Company API keys
  • An open-weight model you host

Security checklist

  1. Nothing ships without a person.

    The rollout waits until someone clicks Allow. The approval lands in the audit trail.

  2. Agents act with the permissions of the person who started them.

    A coding agent never has more access than its developer.

  3. Your code stays in your environment.

    The workspace connects to your Git provider. With a model you host, the code never leaves.

  4. It runs on your Kubernetes.

    Wherever you want, as long as there is Kubernetes. If you do not run Kubernetes yet, we provide it.

04 / Proof

We build Giant Swarm with coding agents.

Our own engineers are the first users. In one night our agents opened 1,000 pull requests. These are the figures from our own work.

  • 300xmore pull requests
  • 17xfewer tool calls
  • 9xfewer tokens
  • 3xlower cost
Giant Swarm allows our small team to offer a world-class platform for all our modern applications and AI workloads.
Simon Gellrich Manager Web Development, EGGER

See it with one of your repositories.

A 30-minute walkthrough on a lab installation takes one small feature from spec to staging and ends with the approval and the bill. The first team starts with the Foundation Package at a fixed price.