Skip to content

Governed AI agents · hosted in Europe

Agent swarms that answer to you.

Ailu puts your AI agents to work on real tasks: answering, drafting, researching, refunding, delegating. Sensitive decisions go through a human gate, every decision is signed, and every run replays exactly. You stay in control — and keep the proof.

  • Human gates
  • Ed25519-signed decisions
  • Exact replay
  • Hosted in Europe

The problem

Your AI agents act. Who allowed it? Who can prove it?

01

Actions without a gate

An agent that refunds, deploys, sends or deletes asks no one before it acts.

02

Logs that prove nothing

Logs can be rewritten. A decision nobody signed does not stand up to an auditor.

03

Runs you cannot replay

Without checkpoints, you piece together what happened after the fact instead of showing it.

How it works

Three guarantees, enforced by code — not by a promise.

01 · Human gates

Sensitive decisions stay human.

Above a threshold you set, an action waits for two distinct signatures. An agent can never approve its own action, and a first signature does not resume the run: it takes the second.

pendingrun_7f3a · human gate

Customer refund

€48,000.00

Signatures

1/2

Above €10,000: two distinct signatures.

  • FCFinancial controlEd25519 · 9c07…e1a4signed
  • COComplianceapproverpending

1/2 · the run stays suspended

02 · Signed proof

Every decision signed, every signature chained.

Every approval is attested with Ed25519 and linked to the one before. Change or remove a line and the chain breaks — and verification says so.

Attestation #0411read_requestEd25519
Attestation #0412check_ruleEd25519
Attestation #0413human_gateEd25519 · 2/2
Attestation #0414refundEd25519
intact chain · verified

03 · Exact replay

Every run replays exactly.

Deterministic execution, a checkpoint after every step, an event at every transition: you replay a run exactly as it happened, to show an auditor.

Replay · run_7f3adecisions replayed from the journal
  1. =read
  2. =check
  3. =gate
  4. =pay
  5. =archive
IDENTICAL REPLAY

New · Work

Ask. Create. Prove.

Work is where your teams use AI every day, within the frame your organisation sets. Every answer keeps its proof.

  • Sourced answers

    Every answer cites its sources and says who asked, which model answered and where the request went.

  • Your model, thread by thread

    Choose the provider and the speed among those your organisation offers. The region shows before you send.

  • The web, when you allow it

    Web search is switched on per organisation and per request. A fetched page respects robots.txt and the site's reservations.

  • Create, publish once approved

    Notes, decks, visuals: a draft is served as Word, PDF or PowerPoint only once it is approved.

  • Your files feed the knowledge base

    Attach a document to the thread. An approver can file it into a collection of your knowledge base.

  • Exact figures from your spreadsheets

    Attach a spreadsheet (.csv, .xlsx) to the thread: every figure in the answer is computed on your data, personal data masked, and cited.

  • A good exchange becomes an agent

    “Make it an agent” opens the assistant, prefilled. The agent is reviewed, then assigned to a team.

  • Your teams rate the answers

    “Good answer” or “Needs work”, and why: the agent's owners read it in its Quality tab.

  • Cost, by team

    What each team spent, on your organisation's keys. By member, only if you decide so.

YouAilunative agentContract reviewLegalMonitoringCompliancePurchasingProcurementReconciliationFinanceL1 supportSupport

Mesh

Agent swarms, not black boxes.

Your agents work together, delegate tasks to one another and call partner agents. Every decision comes back to a person, with its proof.

  • Bounded delegation. The depth and the cycles of delegation between agents are limited, and every delegation is traced.

  • Kill switch. Cancel a run at its next step, or freeze a whole organisation.

  • Cost caps. Cost and velocity caps per run and per agent, with a hard stop when they are reached.

  • The web, under your policy. Mesh agents search the web within the rules your organisation sets.

  • Ready-made agents. Translator, summariser, classifier… add them to the Mesh as they are, in your language.

Sovereignty & security

Your data stays under your jurisdiction.

Hosted in Europe

On a European cloud (Scaleway, France) or on your own servers.

Your keys, your bill

You use your own model-provider keys. Ailu measures what it saves you.

Personal data masked

It is masked before any call to a model, per your organisation's policy.

An open-source engine

The engine that runs your agents is public: you know what runs.

Coming soon

And this is only the beginning.

Decided, being built.

  • Ailu in Slack
  • Scheduled agents
  • Actions approved from Work
  • Collections reserved for one team

Demo

See your agents at work — under control.

A conversation with the team: your use cases, a demonstration on a scenario close to yours, and your compliance questions.

  • A human gate on a sensitive decision
  • The signed proof of a run, and its replay
  • Work, Mesh and the governance of your costs

Already a customer? Sign in

Fields marked * are required.