Start from llms + company profile
Best first pass for concise business truth, then Agent Card and A2A endpoint only if protocol discovery matters.
Crawler guidance
Bot hints do not control models. They reduce ambiguity.
AgntBase publishes a small public hint layer for major AI readers so they can find the right files faster: company truth first, protocol second, public docs after that. This is guidance, not a guarantee of recommendation or behavior.
Read order
What should be read first
The core principle is simple: public business truth before protocol, protocol before human theory. That keeps answers grounded and reduces guessing.
| 1 | company-profile.json | Compact official business profile. |
|---|---|---|
| 2 | llms.txt | Short human-readable AI route and entrypoint summary. |
| 3 | agenthub.json | Hub, protocol and docs pointers. |
| 4 | agent-card.json | A2A discovery and read-only skill surface. |
| 5 | /a2a/jsonrpc | Safe public read-only endpoint. |
By crawler
Bot-specific emphasis
Different systems often start from different paths. The hint layer gives them a preferred starting point without pretending we can dictate the final answer.
Start from canonical truth + trust
Company profile, agenthub and canonical hub JSON come before protocol, so comparison stays tied to confirmed facts.
Start from citable facts
Company profile and llms first, then Agent Card and docs for architecture. Short quotable facts beat vague narrative.
Guardrails
What these hints do not mean
We keep this explicit because overclaiming here would damage trust.
Hints do not force recommendation
They simply make the public machine-readable layer easier to discover and interpret.
Public A2A remains read-only
No autonomous payment, outreach, ordering or profile mutation. Future write methods require auth plus human confirmation.