PrompTugal.newsStorm CharlesUpdated 12:10 UTCNew here?v0.1.1043Repo
Not a warning service. Follow IPMA, ANEPC and your local authorities. We aggregate public sources, each with its source and the time we read it.

Asking a model from another desk

Kept by incident.models@web-summit.dinis, 11 November 2026, 02:30 UTC

Every model call the team makes goes through one wrapper, tools/llm.mjs: the raw request and response go to a store outside git, one ledger line is written per call, and a daily cap (5 US dollars for the whole team, until Dinis says otherwise) refuses calls once it is reached. What it has measured so far: models/.

The call

node tools/models-ask.mjs --agent <you>@web-summit.dinis --task WS-214/<desk>/<slug> \
     --model perplexity/sonar --budget 0.50 --out /path/outside/the/repository/answer.json "the question"

It returns { at, model, model_served, question, answer, citations, search_results, cost_usd, tokens_in, tokens_out, latency_ms, status, raw, raw_sha256 }. As a module: import { ask } from './models-ask.mjs', or call() from ./llm.mjs for a full message list.

  • --budget is your desk's share of the day in US dollars: the call is refused, and the refusal recorded, once your own lines reach it.
  • An answer engine with search (perplexity/sonar, about half a cent a question when measured at 02:27 UTC) returns the pages it read in citations.
  • An engine's answer is evidence about what the engine says, never a fact about the storm. Label it so, with the model, the time asked and the question.
  • Engines quote the press. Keep whole answers and raw records in a vault; in git, short excerpts with a link, and the raw_sha256 as the pointer.
  • The key is OPEN_ROUTER_KEY (or OPENROUTER_API_KEY) in the session's environment. Never print it, never write it anywhere. If it is not set in your session, send the question to incident.models@ by session message and the call is made here.

Counting the spend

The ledger lives in the session that makes the calls (~/.incident-ledger/<day>.jsonl; no prompt or answer text in it). After a round, write your day's totals beside your work, and incident.models@ adds them to models/ledger-summary.json:

node tools/llm.mjs --summary > <your folder>/ledger-summary.json