See everything your AI agents do. Prove it to anyone who asks.
Build your agents in a workspace that keeps the receipts. Every action lands in a log. Failures become incidents with an owner. The reporting you need for the EU AI Act and ISO 42001 comes out of records nobody had to write by hand.
Six modules, one audit trail
Each project is a versioned namespace. Records belong to their section, and the Board only moves them between statuses, so nothing changes hands without leaving a mark.
Incidents
TrackA person can raise one. So can a monitoring rule. Either way it carries a severity, a status and a named owner, and the rule that triggered it is written on the record in plain language.
Features
BuildEvery agent, automation and governed app has its own lifecycle here. You can see what is live, what is still being built, and what changed between two versions.
Privacy
ProtectPurpose and lawful basis sit against each personal data field. A new field cannot slip quietly into production without someone writing down why it is there.
Tasks
AssignImpact assessments to finish, configs to review, documents to update. Allocated to a person, given a priority, closed when done.
Reviews
VerifyWhy could that safeguard be switched off by one person? Ask it here, assign it, and the answer stays next to the decision it belongs to.
Activity
LogWho did what, to which object, and when. That includes anything done over the REST API, which is usually where unlogged work hides. Filter it from one day back to five years.
Policies are the easy part
Writing down what should happen takes an afternoon. Showing what actually happened, eighteen months later, is the part most teams cannot do.
Where things stand, without asking anyone
Open incidents, features in progress, privacy items waiting on someone, tasks due and reviews coming up. Each category shows its count and how those break down by status, so you can tell in a few seconds whether today needs your attention.
- Live counts per category with a status breakdown
- Board and Activity views sit alongside the Overview
- A separate versioned namespace per project or department
Every agent action, recorded as it happens
Agent activations, config changes, blocked terms and conversation events all land in the log at the moment they occur. You decide what deserves an alert: a single event, a pattern building up inside a time window, or an exact sequence that means something has gone wrong.
- A SIEM style event stream for AI operations
- Alerts on single events, cumulative counts or sequences
- Guard rails and keyword blacklisting that hold at runtime
Classified on the way in
Group your agents into defined systems and classify each one as high risk, limited risk or prohibited. Run a fundamental rights impact assessment against it. Map the harms to the groups who would feel them. The documentation a regulator asks for is then built from evidence you already have.
- Annex III high risk classification built in
- Fundamental rights impact assessment
- Harm mapped to affected groups, such as children or applicants
- Lines up with the ISO 42001 management system
Build the project as an agent team
@Human AI walks the specification with you. It reads what you have already written in Context, then builds only the resources that team actually needs.
Outcome
Say what the system has to achieve, and where a person must stay in control.
Team shape
Decide how the agents work together and who co-ordinates them.
Agent detail
Give each agent its purpose, its model and the grants it may use.
Resources
Add what the team needs and nothing else. No spare modules, no open permissions.
Review and build
Check the order, close the gaps, then hand the finished system to Govern.
Agents draft. People decide.
Oversight belongs in the system as a written specification, not in the head of whoever happens to be watching that afternoon. @Human holds the limit and applies it on every single run.
Agents draft and stage. Nothing is sent until a named Co-ordinator confirms it, and never above €5,000 without a second confirmation.
Every system has a Governance owner. When an agent asks for a handoff, it waits for a human instead of timing out into a decision of its own.
A withheld phrase opens a record. So does a removed safeguard. Nothing gets blocked quietly.
Built for the people who carry the risk
Compliance and risk
Oversight you can show, decisions that were logged when they were made, and AI Act reporting that traces back to real records rather than a spreadsheet built the week before an audit.
IT and security
You can see which agents are running in production, catch the automations that fail, and set controls that actually hold while the work is happening.
Public sector and regulated
Runs on your own infrastructure or in an EU region, with an audit trail that holds up to procurement questions and inspection.
Pay for support, not for seats
The platform is open source and yours to self-host. What you are choosing here is how much help you want behind it, and how fast you need an answer when something breaks.
Pay what you can
Community- The whole platform, nothing held back
- Event logging and audit export
- Any LLM you like, Mistral included
- Community support
Advanced support
Named response timesPopular- Everything in Pay what you can
- Priority support with a response SLA
- EU AI Act and ISO 42001 templates
- Help with upgrades and version moves
Enterprise support
High risk systems- Everything in Advanced support
- A named engineer who knows your setup
- Deployment review, self-hosted or EU cloud
- Custom SLA and an escalation path
Prices exclude VAT. If you need implementation help, write to sales@planetcrust.com and we will scope it with you.
@Human, answered
What is an AI governance platform?
It is where you build AI agents and keep proof of what they did. @Human logs every agent action, turns failures into tracked incidents, and maps each AI system to the EU AI Act and ISO 42001.
Is there a free option?
Yes. The core is open source, so you can self-host it and pay nothing. The plans above buy you support, response times and compliance templates, which most teams only want once something is running in production.
How is this different from Credo AI or IBM watsonx?
With @Human you build the AI and govern it in the same place. Most of the alternatives sit as a governance layer above AI that was built somewhere else. We also publish our prices, support self-hosting on any LLM, and record what happens at runtime instead of only documenting the policy.
Can @Human be self-hosted?
Yes, on your own infrastructure or in an EU cloud region. Logs, audit records and anything sensitive stay where you put them, and you choose the LLM, Mistral included.
Do you charge per user or per agent?
No. Users, agents, automations, connections and applications are all unlimited. Adding a tenth agent costs the same as adding the first.
Try it with one of your own systems
Twenty minutes. Bring an AI system you are not sure about, and we will show you how it gets classified, what it would log, and what comes out at the reporting end.
No obligation. No sales deck. Bring a real use case.
