Architecture, with evidence
Your AI is the interface. NitroBot is the part that has to be right.
How scoped tools, deterministic engines and reviewed commands turn a bot’s proposal into accountable work.
A conversation becomes a checked command
Your AI interprets the request and calls scoped tools through the Model Context Protocol (MCP). NitroBot checks the structured input, evaluates the business rules and returns a result the bot can explain. A convincing answer alone cannot book a resource or authorize an effect.
The application command owns the write. Before a new commitment, it checks current authority, the exact reviewed proposal and the records that proposal depended on. Human review or a previously granted, bounded standing authorization provides permission to proceed.
- Your AI
- MCP
- NitroBot commands
- Pure engines
Put accuracy in a rerunnable engine
Pure engines receive explicit inputs and return decisions without credentials, database access or provider calls. A scheduler can therefore check capacity and fit using the same rules in a test, an operational host or an isolated simulation.
The source and test links below each explanation let you inspect the mechanism. They establish the scope exercised by repository tests; they are not evidence of a production deployment or a live provider result.