AI agents that take real operations work off your team — installed with guards, and run for you.

Business consulting, research, data analysis and digital tools for companies, delivered as agentic operating systems: AI agents, memory, guards and workflows that run a company's work.

For founders and COOs of a 10–200 person EU company who carry the operations load personally.

Book a scoping call
agent 1 agent 2 agent 3 boot guards shared memory context loaded on boot read / write

what we install

Company Brain

A team knowledge base your agents can read and write, installed with boot, handoffs and guards.

Fixed-scope installfrom €10,000

Agentic Readiness Audit

Two weeks: where agents can safely take over your work, ranked, with one working pilot.

Fixed pricefrom €2,500

Signed Agent Identity

Make your company's agents identifiable and welcome on the web: signing keys, a transparency page and verified-bot submission.

Fixed pricefrom €1,500

Agent Operations

We run your agent team: monitoring, repair and new workflows.

Monthly retainerpriced per scope

How we run our own company on agents

Everything below is live and checkable. Our own company runs on it.

  • agent.leematulis.com — live, with a signed Web Bot Auth key directory and a transparency page documenting how our agents behave.
  • Every request is signed with HTTP Message Signatures (Web Bot Auth); the public keys are published at /.well-known/http-message-signatures-directory.
  • Our own agents made 1,518 web fetches between 25 August and 24 September 2026 — 11.7% were blocked.

behaviour rules

  • Follows robots.txt rules for its own user agent.
  • Makes at most one request per second to any one site.
  • Never logs in, fills in a form, buys anything or solves a CAPTCHA.

the guard patterns behind it

Red-first
A safety check counts as installed only after we have seen it fail on the exact problem it exists to catch.
Claim guard
Two agents can never edit the same file at once; each claims its files before writing, and the system refuses a write against someone else’s claim.
Safety guard
Destructive commands are blocked by the system itself, not by an instruction the agent could forget.
Truth guard
Any live number on a page is computed from its source, never typed by hand.

This page is also written for agents: machine-readable facts live at /llms.txt, and an MCP server with tools agents can use for their owners runs at /mcp.

Founded by Lee Matulis

Book a scoping call