Question
What colour is the cat?
Answer in the test context
Petrol.The direct rule.
AI in your company context
Your AI should know what applies here.
PetrolCat brings your organisation’s knowledge, policies and decisions together. People and connected AI agents work from the same traceable foundation. Reviewed work can be used again on the next task.
For companies, public bodies and teams working under binding rules.
The origin
Why PetrolCat?
It started with an unusual test: could a language model apply a deliberately defined rule rather than guess from general world knowledge?
The rule in our test worldCats are petrol.
Question
What colour are its kittens?
Answer in the test context
Petrol.The rule carries through to the offspring.
Question
What colour is a white tiger?
Answer in the test context
Two statements now conflict.Its place in the cat family and its explicitly stated colour point to different answers.
The third step mattered most.
In the original experiment, the rule led the model to say “petrol” even for the white tiger. That exposed the real product question: when does a rule apply, which exception takes precedence, and when must AI flag a conflict? A sound system must not present that tension as an unquestioned fact.
That test became the starting point for PetrolCat: AI should work within the right context and recognise where that context reaches its limits.
The product
Turn an answer into a reliable foundation.
A rule in a prompt is not enough. What matters is who sets it, where it applies, what supports it and how a change takes effect.
Bring in knowledge
Record sources, policies and decisions with their scope.
Work in context
People and connected agents retrieve relevant information for a specific task.
Review and decide
Responsible people resolve conflicts and approve sound outcomes.
Use it again
The reviewed foundation remains available with its source and context for future work.
Available today
Coding agents can take part, too.
Through MCP, connected agents can retrieve knowledge and applicable decisions, submit planned actions for assessment and propose new findings. They can keep working in their team’s tools while bringing PetrolCat into the workflow.
MCP makes assessment available. It does not automatically stop an action outside that workflow.
An agent plans a data export.MCP
- 01 Request
Which policies apply to this export?
- 02 Context
The relevant policy and an earlier decision arrive with their sources.
- 03 Assessment
The agent submits its planned action. The result shows what still needs resolution.
- 04 Feedback
A new finding is proposed and joins the foundation only after review.
The value
Use what you have learned instead of starting over.
Research, clarification and review take time and model compute. PetrolCat keeps reviewed outcomes with their sources. On the next task, the team can build on them and check what still applies. This may reduce repeated model work; the actual effect depends on the use case.
The platform vision
One shared context. Several ways in.
At its centre is a traceable foundation for knowledge, policies and decisions. Additional access points will bring it to the places where people and AI work.
Platform core
Knowledge, policies, decisions and review. Web interface and MCP integration.
Desktop control
Planned local component for added control on the device. Technical enforcement requires the right integration.
Mobile access
Planned app for controlled knowledge access, notifications and decisions by authorised people.
What PetrolCat does and does not do.
Does PetrolCat prevent every hallucination?
No. PetrolCat supplies traceable organisational context and can expose conflicts. Answers and actions still need review appropriate to their risk.
Who decides which policy applies?
People with the appropriate responsibility. The system records the source, scope and decision and makes them usable for later tasks.
Does the MCP integration stop every agent?
No. An agent can ask for policies and submit actions. Technical blocking requires control of the actual execution path.
Next step
Bring a real workflow.
Together, we can examine which sources and policies your AI work needs and where a person must make the decision.
Contact us through the current website