How to wire a strategy / AI agent

Step 1 — Read the index first

Point the agent at /llms.txt — the machine-readable index of datasets, rules and OpenAPI links. Fetch it once, then work only from the dataset ids it lists.

curl "https://twmarketdata.com/llms.txt"

Step 2 — Call datasets with your key

Every data call carries X-API-Key. The envelope is { rows / items, count }; read count before touching the rows.

Step 3 — Trust an empty result

When a range has no data, the API returns an honest empty result — not an error, not a guess. This is a real response:

{  "dataset": "institutional_flow",  "rows": [],  "count": 0}

The rule for the agent

• count == 0 means CONFIRMED no data in range. The agent must say "no data in that range" — never invent a number to fill the gap.
• This is the single most important anti-hallucination check: read count, then decide.

Step 4 — Know the permanent gaps

Some data is deliberately not offered (copyright, no source). A /meta/boundaries endpoint that declares these permanent gaps is in preview and not live yet — until it ships, treat the not_available list in /llms.txt as the source of truth for what is intentionally absent.

Disclaimer

TW Market Data provides data only — not investment advice, not stock recommendations, not trading signals. An agent built on it must not present its output as any of those.