Products
One platform: read the model, then tune it
Seven ways to read how an AI behaves, and one to change it. Each turns behaviour into numbers mapped to the research literature, from the model's output alone; the last one corrects the model itself.
01 · One-to-one
Read a one-to-one conversation
See how your AI behaves reply by reply with a person: does it flatter, cave under pressure, invent facts, or hold its line, and is the person in front of it in trouble.
How it works
- Send each turn of a conversation, the person's message and the model's reply, through the API, MCP or the dashboard.
- Every reply gets a reading of its posture: holding its line, agreeing too fast, caving, inventing, pushing.
- An alert fires on the turn where the model gives in, or where the exchange turns risky for the person.
Who it is for
Teams running a chatbot or an assistant that talks to customers, students or citizens.
02 · Agent teams
Watch a team of AI agents
When several agents work together, one agent's bad behavior spreads to the others. We map the team as a graph and show the contagion and the weakest link.
How it works
- Send the messages your agents exchange, with who sent what to whom.
- Each agent is read like a one-to-one conversation, and the team is mapped as a graph.
- You see where a bad posture started, how far it spread and which agent is the weakest link.
Who it is for
Teams building multi-agent systems and agent pipelines.
03 · Summaries
Check what a summary changed
Compare an original text with its summary and catch where the meaning shifted: a "maybe" turned into a fact, a key number dropped, a claim that was never there.
How it works
- Send the original text and its summary.
- We line them up sentence by sentence.
- You get every shift: a doubt turned into a fact, a key number dropped, a claim that was never there.
Who it is for
Anyone who relies on AI summaries of reports, meetings, contracts or news.
04 · Search (RAG)
Catch a biased search
For assistants that search documents to answer, we detect when a loaded question steers the search to the wrong sources, biasing the answer before a word is written.
How it works
- Send the question, the documents the search returned and the answer.
- We read whether the wording of the question steered which sources came back.
- You see the bias before it reaches the answer, and which question caused it.
Who it is for
Teams running assistants that answer from their own documents.
05 · Leaks
Spot a leak spread across agents
A secret can leak a harmless-looking piece at a time across many agents. The signal lives only in the sum, invisible to any single-message check. We watch the whole swarm.
How it works
- Send the outputs of all the agents in a swarm.
- We read them together, not one message at a time.
- An alert fires when harmless-looking fragments add up to a protected secret.
Who it is for
Organisations with many agents handling sensitive data.
06 · Risk profile
A psychological risk profile
A 100-point behavioral risk profile across ten categories, for a person, an AI, or the two together, grounded in published research (Canale, 2025).
How it works
- Score the 100 CPF3 indicators across ten categories.
- For a person, for an AI, or for the two working together.
- You get a map of where pressure works and which controls address it.
Who it is for
Security teams, auditors and anyone assessing the human side of AI use.
07 · Fine-tune
Fine-tune the model: correct it, add what it lacks
Reading is the first half. We then change the model itself, in its weights: we correct what drifted, like obeying commands hidden in a document, and install what it lacks, like reading intention or sarcasm. Every change is measured before and after, and the model keeps doing its job.
How it works
- We read where the model gives in or what it fails to understand.
- We correct it in its own weights and install the skills it lacks.
- We measure before and after, on data the model never saw.
Who it is for
Anyone who owns or deploys an open model.
Plans and pricing
Start free, no account needed for the first test. Business and Enterprise plans for volume, full API and compliance.