For your AI assistant
Let your assistant do the vendor legwork.
Evaluating Signal is a reading job — samples, provenance dossiers, pricing, docs. Your AI assistant can do it in minutes and show its work. Paste one prompt; get an internal memo with every claim cited and everything unverifiable flagged.
Every page the prompts reference is public and keyless — and our robots.txt explicitly welcomes assistant fetchers on these paths. Verify, don’t trust: the memo cites its sources.
Paste into Claude or ChatGPT
An evaluation memo, from one paste
Your assistant fetches a pack sample and its provenance dossier, spot-checks the corpus over the keyless lookup, reads pricing and the docs index, then writes a one-page internal memo — citing the page behind every claim and flagging anything that did not check out.
I'm evaluating Uphealth Signal (uphealth.us) — a cued-message API for patient adherence programs — as a potential vendor. Do the evaluation legwork from its public pages, and verify rather than trust: 1. Fetch https://uphealth.us/signal/packs/diabetes-pharma-init/sample and https://uphealth.us/signal/packs/diabetes-pharma-init/dossier — real sequence messages with their federal sourcing. 2. Spot-check the content is real: fetch https://api.uphealth.us/v1/public/health-fact?q=diabetes-basics (free, keyless JSON) and confirm the cited .gov source actually supports the message. 3. Read https://uphealth.us/pricing and https://uphealth.us/llms.txt — tiers, Display vs Deliver modes, BAA availability, and the API surface. Then write a one-page internal evaluation memo: what the product does, content provenance, integration lift, pricing and compliance posture, and open questions for a vendor call. Cite the uphealth.us page for every claim, and flag anything that didn't check out.
Works in any assistant with web browsing. Each pack sample page carries a pack-specific version.
Paste into Claude Code or Cursor
Integrate in an afternoon
For your engineer or solutions team: a coding agent reads the integration guides, and once a human pastes a free sandbox key it runs a five-day sequence end-to-end — then reports every response shape it saw and its own estimate of the production lift.
Integrate Uphealth Signal — a cued-message API for patient sequences — into a scratch project. 1. Read https://uphealth.us/agents.md end-to-end first: it is the single build recipe (auth, create a stream, the daily cue loop, the response shape, going live), and it points to the OpenAPI spec and llms.txt for anything deeper. 2. If you support MCP, add the server: claude mcp add --transport http uphealth https://mcp.uphealth.us/mcp — its keyless lookup_health_fact tool needs no key. 3. I'll paste a free sandbox key (up_sandbox_...) from https://uphealth.us/signup — don't create accounts or keys yourself. The sandbox is Display mode: 50 cue calls a month, no PHI. 4. Following agents.md, create a stream from a live template, then run the cue loop for a 5-day sequence: submit a plausible response_action each day (include at least one no_response), pass an idempotency_key per response, and note the receptivity float and safety verdict on each cue. 5. Finish with a write-up: every response shape you saw — create, cue, _meta, and any errors (a /cue without feedback should 409) — plus your estimate of integration effort for production.
The agent never creates accounts or keys — you paste the sandbox key yourself. Sandbox is Display mode: no PHI, no BAA, 50 cue calls a month.
Writing code?
The whole integration in one file.
Point your coding agent — or yourself — at a single build recipe: auth, a first cue, the daily loop, the response shape, and going live, with the OpenAPI spec one link away. Read it top-to-bottom and scaffold the integration in one pass.
The reading list
What your assistant will read
- agents.md — the single build recipe a coding agent reads top-to-bottom to scaffold the integration in one pass.
- llms.txt — the machine-readable index of the whole developer surface, packs and pricing included.
- Pack samples — real sequence messages in arc order, with the federal agency behind each one.
- Provenance dossiers — message-by-message sourcing: the .gov link and the quoted line each message rests on.
- The keyless lookup — one federal fact as JSON, free, no key: live proof the corpus is real.
- Pricing — the four Signal tiers, Display vs Deliver, and where a BAA applies.
- The docs — the API reference and guides a coding agent builds against.