Skip to content

Services

Agents that do real work.

Agents that handle routine work and check with a person before anything irreversible. Assistants in Signal or Slack, support resolvers, contract review and coding agents on private repos.
Book a scoping callSee the Run retainer

Who it is for

  • Teams buried in manual workflows

    Routine work that follows the same steps takes hours your team could spend elsewhere.

  • Founders who need leverage without headcount

    Agents take on the routine work so a small team can do more.

  • Organisations that cannot send data to third-party clouds

    We favour deployments where your data and keys never leave your infrastructure.

What you get

We embed with your team, map your real workflows and data, then design and ship custom single or multi-agent systems built around the models that perform best for your use case.

  • Assistants, support, document and coding agents

    Assistants in Signal or Slack, support resolvers, contract and document review, and coding agents on private repos.

  • Architecture around your workflows

    Orchestration, memory and tool use designed for the way your team already works.

  • Models benchmarked on your tasks

    Model selection tested on your actual tasks. Open-weight first when it wins.

  • Local-first deployment

    Wherever it is workable, your data and keys stay on your own infrastructure.

  • Guardrails and human approval

    Guardrails, an evaluation harness and human-in-the-loop controls. Payments, outgoing messages, deletions and signatures wait for a person to approve them.

  • AI Agent OpSec Review

    A one-off add-on: prompt-injection surface, tool permissions, memory and data-flow audit, with scored findings.

  • Documentation and handover

    Your team owns the system.

How it works

  1. 1

    Scoping call

    We map workflows, data sensitivity and success criteria.

  2. 2

    Architecture and prototype

    A working vertical slice on your real data.

  3. 3

    Ship and harden

    Production deployment, evals, OpSec review and handover.

A live example: Hydra

A multi-agent system we built and run on mocked company data.

Your company's new AI colleague, ready to work.

Mail, calendar, tasks, treasury and invoices in one workspace anyone can use in plain English. It processes invoices on arrival, ranks them by urgency and keeps your team focused on the important deals.

  • Inbox to invoice
  • Read-only by default
  • Runs on a schedule

Built on mocked company data. Your deployment runs on your real systems.

Custom and packaged agents

Have one built for your workflows, or start with an agent we already run.

  • Custom agent build

    A single or multi-agent system designed, shipped and handed over to your team.

    Book a scoping call

  • AI Agent OpSec Review

    A one-off review of an agent you already run: prompt-injection surface, tool permissions, memory and data flows, with scored findings.

    Request a review

  • Counterparty Watch

    A weekly delta brief on the wallets, protocols and entities you designate, with diff-based alerts only when something changes.

    Set up a watch

  • Intelligence feed

    Continuous monitoring briefs on the protocols and entities you care about, built on the same pipeline as our public IntelHub feed.

    See the IntelHub feed

  • Data-broker scrub

    A machine scan of data-broker and directory listings, then a person files your opt-outs and erasures. The watch keeps automated checks and re-scrubs running. Prices and terms are on the privacy page.

    See privacy prices

Keep it running

After handover, the Run retainer gives you reserved engineering hours, model and tooling watch, security reviews of agent permissions and data flows, and a quarterly architecture review.

See the Run retainer

Questions

Will an agent act without asking?
Not on anything irreversible. Payments, outgoing messages, deletions and signatures wait for a person to approve them.
Where do the agents run?
Local-first wherever it is workable. We favour deployments where your data and keys never leave your infrastructure.
Which models do you use?
The ones that perform best on your actual tasks, benchmarked before we build. Open-weight first when it wins.
Who owns the system?
You do. We document it and hand it over. If you want us to keep operating and extending it, that is the Run retainer.
What does the OpSec review cover?
Prompt-injection surface, tool permissions, memory and data flows. You get scored findings.

Book a scoping call.

We map your workflows, data sensitivity and success criteria, then build a working slice on your real data.
Book a scoping callSee the Run retainer