​
Model endpoint

Any server that speaks the OpenAI chat-completions API: OpenAI, Azure OpenAI, OpenRouter, Groq, Together, a local Ollama or vLLM. Set the base URL (there is no default endpoint; the resolved /chat/completions address is shown under it), the model name the server expects and the API key, sent as a bearer token. Until all three are set the model is not configured and every job that needs it says so. Test the connection makes one small call and reports the model, time, tokens and cost. Input price and output price, in USD per million tokens, are used to estimate the cost of each call when the server does not report one. OpenRouter reports exact costs, so there they are only a fallback. Leave both empty to record tokens only. The model is used for AI research (choosing pages after a scan, analysing policy, drafting playbooks and reviewing changes) and for the Executor Kit’s Draft with AI button in the client app, which drafts the digital assets letter, the vault access instructions and the final wishes from the client’s own inventory and vault. Responses are requested as JSON at low temperature. Nothing the model writes reaches a case until an operator applies it, and a kit draft is only ever text in the client’s editor until they save it. The text of every prompt is on the Prompts tab.

​
Usage and cost

Every model call is recorded with its tokens, latency, attempts and cost. Choose a period (the last 7, 30, 90 or 365 days, or all time) to see the totals: calls and failures, tokens in and out, cost (reported, estimated or both) and median and p95 latency; then the breakdown by job (research kinds and each Executor Kit section) and by model, and the recent calls with their result. A call that had to be retried without a request parameter the server refused (a temperature, JSON mode, the usage flag) carries an amber badge naming the parameter, and one whose JSON came from a reasoning field rather than the message says json from reasoning, so nothing the client adapted to is hidden. Each research card on a provider carries its own tokens, cost and time as well.