AgntBaseStack
Project direction

Short map of where AgntBase is going

We are not turning the site into a random collection of AI files. The path is deliberate: make businesses readable, then discoverable, then trustworthy, and only later actionable under controlled rules.

Now

What exists today

The foundation is already broader than a file package: canonical truth, registry routes, handoff logic, diagnostics and a first agent-facing protocol surface.

01

Readable business layer

Canonical profile, company-profile.json, llms.txt, structured data and clear business facts.

02

Registry layer

Hub profiles, vertical discovery, profile quality controls and noindex-safe public structure.

03

Protocol layer

Agent Card, A2A JSON-RPC, read-only search, compare, trust explanation, site check and checkout instructions.

Next

What comes next

We keep adding useful layers only when they improve real discovery, trust and handoff — not just because the standard allows it.

Next 7 days
  • Strengthen weak hub profiles.
  • Add A2A metrics to the admin dashboard.
  • Keep bot-specific hints and agent sitemap up to date.
  • Improve checkout and package explanation language.
Next 30 days
  • Signed Agent Card.
  • Partner API keys and rate tiers.
  • Extended Agent Card for gated metadata.
  • Task-style A2A methods for longer checks.
Guardrails

What we are not doing yet

This part matters as much as the roadmap. We do not pretend autonomy where safe merchant control is still required.

No fake autonomy

No public write methods

Public A2A stays read-only. No profile mutation, no automated ordering, no outreach sending.

Merchant-owned next step

Checkout stays merchant-controlled

We explain package and checkout paths, but do not take money or create business actions automatically through the public endpoint.

Hard rule

Auth plus human confirmation

Any future write-capability will require explicit auth, scoped access and human confirmation with logs.