Skip to content
iftodo
Live agent, not a recording

Hand it a job you're sick of.

A live agent scopes the one we'd build to run it: the frontier kind, built to survive production, not just demo it.

Live, unscripted

Talk to it

iftodo

I'm iftodo's agent, and I'm the real thing, not a recording. So tell me a job you're sick of, the kind that quietly eats your team's week, and I'll do a piece of it right here.

or start with one of these

Enter to send · Shift+Enter for a new line · Paste/drop an image or text PDF

or book a call

You're chatting with an AI agent. Conversations are stored to improve it. See our Privacy Policy.

EU-hostedYou own the codePay only when it survivesHuman approval gate

How far it goes

The work decides the level.

Some jobs need a hand inside one step. Some should run themselves. We build whichever the work calls for, and nothing heavier.

  1. 01

    Assistant

    Drafts and decides inside a step. You stay in control of every call.

    For instance

    Reads every support email and writes the reply. You just hit send.

  2. 02

    Autonomous

    Runs the whole function unattended. Escalates only what needs a human.

    For instance

    Onboards each new customer across your stack, and pings you only when something is genuinely off.

  3. 03

    Self-learning

    Sharpens as it runs. The exceptions you resolve become rules it follows next time, with your sign-off.

    For instance

    Resolve an edge case once and the same thing stops landing on your desk.

At the top end, a whole function runs itself. The same autonomy behind our National Dutch AI Cup 2026 finalist agent, now pointed at an entire finance back office, a 24/7 ops control tower, or just your boring week.

proof, not demos

National Dutch AI Cup 2026 finalist

We already ship the crazy ones.

Behind iftodo is a TU Delft AI engineer who ships real autonomous agents in the open. You work with the builder, not a sales desk.

  • Autonomous researchresearch-copilot
    National Dutch AI Cup 2026 finalist

    An autonomous research agent that runs the whole experiment loop end to end.

  • Agent infrastructureTrustChain

    Decentralized trust for AI agents: a Rust core, QUIC P2P, an MCP server, and pip/npm SDKs, on TU Delft research. The public tip of a larger multi-agent system I'm building, most of it still private.

The bigger system, an LLM agent economy with its own trust layer, stays private for now. Happy to walk you through it on a call.

more on GitHub

Why it holds

Spinning up an agent is easy now. Keeping one alive in production is the whole job.

Survives the mess

Real inputs, real exceptions, the cases that quietly break a weekend demo.

Wired into your stack

It acts in the tools you already use, with your approvals in the loop.

Held up over time

Monitored and corrected as your tools and edge cases shift. Shipped to keep working, not shipped and abandoned.

  • EU-hosted infrastructure
  • You own the code
  • Built for EU AI Act transparency
  • Your approvals in the loop

How we work

From a call to an agent that holds.

No long proposal phase. We scope on a call, fix the price, and build against your real inputs. You pay the rest only once it survives them.

  1. 01

    Scope call

    A 30 minute call. We map the workflow, the inputs, and where a human gate belongs. You leave with a fixed price, not an estimate.

  2. 02

    Build on real inputs

    We build against your actual data and tools, not a happy-path demo, with your approvals wired into the loop.

  3. 03

    Survive the mess

    You sign off only when it holds against your real inputs, exceptions and all. Half up front, the balance on survival.

  4. 04

    Stay alive

    Optional Agent Ops keeps it monitored and corrected as your tools and edge cases shift. Cancel monthly.

If a build cannot be made to hold against your real inputs, you do not pay the balance.

Questions

Before you book.

What stops it from doing something wrong?
You set where a human gate sits. The agent pauses for approval on the calls that matter, logs every action, and never acts outside the tools and limits you grant it.
Which tools does it work with?
The ones you already use. We build against your real stack, your inbox, CRM, helpdesk, files, and APIs, instead of forcing a new platform on you.
What about my data?
It stays in the EU. We prefer EU-hosted infrastructure, keep a full action log, and you own the code and the agent at the end.
How long does the first build take?
Two to four weeks for a Pilot, one to two for a Rescue. You get a fixed price at scoping, before any work starts.
What if it does not work?
On a build you pay half to start and the balance only when it survives your real inputs. If it cannot be made to hold, you do not pay the rest.
Who actually builds it?
One builder, a TU Delft AI engineer and National Dutch AI Cup 2026 finalist. You work with them directly, not a sales desk.

Pricing

Build only when it is worth it.

Three ways in. We scope it, build it against your real inputs, and hand it over documented. On a build you pay half to start and the balance only when it survives your real inputs. You own the code, and we stay only while the agent is carrying real work.

Book a callor talk to the agent

  • first build

    Pilot

    from EUR 3,500

    Your first agent, built against your real inputs in 2-4 weeks.

    Price fixed at scoping. You own the code. 50% to start, the rest only when it survives your real inputs.

  • stalled pilots

    Rescue

    from EUR 2,000

    Built one that broke? Bring it. Audit, hardening, and a clean handover.

    1-2 weeks. Honest verdict first: if it is not worth saving, we say so.

  • keep it alive

    Agent Ops

    EUR 500-2,500 / mo

    Monitoring, evals, and drift fixes, so it keeps working as your tools change.

    Scales with volume and criticality. Monthly report. Cancel monthly.