Skip to content

AI agents

AI agents that do real work, within limits you set.

An agent handles work that needs judgment across several tools. We agree with you on what it may do and where a person approves, then measure it on your real tasks.

Who it's for

Work that needs judgment across several tools, such as triaging an inbox or ticket queue, researching and drafting, processing documents or an internal assistant that takes actions.

What we deliver

  • An agent design that lists its tools, its permissions and where a person approves.
  • A build with the Claude Agent SDK, or with n8n agents for simpler cases.
  • Guardrails: allowed actions only, spending limits and full logs.
  • A test set of real tasks, so you can see how often it gets them right.
  • Deployment in your own cloud when data must stay there.

How we measure it

  • Task success rate on the test set.
  • How often it escalates to a person.
  • Cost per task and time per task.

Example

The skill executor project below shows one pattern: a Claude Skill wrapped in a small web app that runs it on a server. Users upload a file, click Run and download the result. Risky shell commands are blocked before they run, and each run reports its cost.

Related work

Built in the open

Questions

Common questions

Can an AI agent do things without our approval?
Only what you agree to. Agents get the least access they need and every action is logged. Sensitive steps wait for a person.
Which AI models do you use?
Claude, OpenAI models and open-source models, chosen per task. The test set decides, including cost.
Where does our data go?
It stays in your accounts wherever possible. AI models are used through business API plans whose terms don't allow training on your data, or through your own cloud (for example, AWS Bedrock) when it must stay there. [TODO: verify current provider terms before publishing]

Have a process that eats hours every week? Let's look at it together.