Stated plainly, because you are the one who has to stand over it in front of your client.
Watch
LiveFits: Regulated clients, and anyone whose customer has a security questionnaire
Grades a public MCP server A to F against the OWASP Agentic Top 10 and NIST IR 8596, and goes deeper under signed authorisation. The output is a report your client forwards to their own customer's security team. You cannot self-certify your own build, which is the point: the independence is the product.
See Watch→Cost
BetaFits: Clients on tokens, spending roughly EUR 2,000 a month or more
Wraps their Anthropic, OpenAI or Gemini client with one line per route and attributes every euro to a route, feature or agent. Then it replays their real traffic against a cheaper model and only recommends the switch once an independent judge agrees on 95% of answers. It recommends; a person applies it.
See Cost→Ledger
Being builtFits: Clients on seats, where the buyer is not technical
A monthly report read from the client's Microsoft 365 tenant under admin-granted read access. It reconstructs what the AI actually produced, attributes it to projects, and reports volume, time displaced and unit cost. It is the answer to "what did we get for all this" from someone who will never open a dashboard. Not finished, and not sellable yet.
Eval
In developmentFits: Clients where a wrong answer costs something
Turns a system prompt into a test suite, then scores real production output against it daily and reports which check started failing. It needs access to output the agent already saved, not the API path, so it reaches clients Cost cannot. None of it is built.
See Eval→