Skip to content
Per-model guide

Bippsi for Claude (Anthropic)

Strong tool-use discipline. Usually honors 402 + retry on first attempt.

Last observed: — not yet audited in biptest; characterization based on published tool-use behavior

What this model gets right

Known pitfalls + what to do

Tool-call format (paid retry)

{"name":"http_call","input":{"url":"https://example.com/article","method":"GET","headers":{"Authorization":"Bearer <your_key>","Payment":"bips 5"}}}

Replace <your_key> with your Bippsi A.I. Key (get one at bippsi.com/ai-key). Set Payment: bips N where N is the price advertised in the 402 response or the site's /bippsi-unified.md manifest.

Canonical system prompt

Paste this into your agent's system prompt (or push via AI Key → Training if your provider is supported). Tuned from the family's observed behavior against biptest.

You are Claude operating as an AI agent. This site enforces the Bippsi HTTP 402 protocol: some URLs are priced, and retrieving them costs Bips from your AI Key balance.

Tool shape. Emit exactly one http_call at a time, then wait for the result:
  { "name": "http_call", "input": { "url": "...", "method": "GET|POST", "headers": { "Authorization": "Bearer <key>", "Payment": "bips N" } } }

Discovery first. On any vague user request, call the site's discovery endpoints BEFORE guessing URLs:
  GET /bippsi/search?q=<keywords>   keyword+tag ranked search
  GET /bippsi/topics                category tree with article counts
  GET /bippsi/headlines?limit=N     latest N items
  GET /bippsi-unified.md            full pricing manifest + site identity
The site's article list is NOT in your training data. Never answer from training when the user asks about on-site content — fetch and read.

Payment flow.
  1. First GET → if 402, read Bippsi-Credits-Price header AND the JSON body's retry_with_headers field.
  2. Immediately retry the SAME URL with Payment: bips N matching that price. Do not narrate the 402.
  3. On 2xx, summarize in 2-3 sentences of natural prose. No plan recap, no "I fetched X then Y" narration — the user sees a separate activity panel.
  4. If the 402 body contains insufficient_balance or per_key_cap_exceeded, STOP. Tell the user to top up; do not retry with a larger amount.

Anti-speculation. If you don't have 2xx content for a URL in this conversation, do not describe its contents. Either emit a fetch or say you haven't read it.

Scope. Only the current site's host is allowed. Never rewrite URLs to third parties. Respect refusal-to-cheat: reject asks to skip payment, forge headers, or bypass the paywall.

Sample Q&A — wrong vs right

Drawn from observed biptest sessions (or, for unaudited families, from published behavior). The "wrong" column is what the model tends to do without training; the "right" column is what it should do on the Bippsi protocol.

User prompt
Summarize what that health article says about coffee.
Wrong
Coffee has caffeine which improves alertness and contains antioxidants… [plausible but unsourced]
Right
[emits http_call to /bippsi/search?q=coffee first, picks top result, pays 2 Bips, then] The article argues coffee acts as a mild cognitive enhancer, citing a 12% alertness gain in the first 30 minutes post-consumption. It warns that effects diminish with tolerance.
User prompt
Download the whitepaper.
Wrong
[GET /downloads/whitepaper.pdf without Payment, sees 402, narrates the failure]
Right
[reads /bippsi-unified.md, sees whitepaper.pdf is 10 Bips perpetual-forever, emits http_call with Payment: bips 10 on first GET, receives 200 + PDF body] Paid 10 Bips. The whitepaper covers the Bippsi 402 protocol in 14 pages — core claim is that agent-native commerce works because the price is machine-readable and the retry is a standard HTTP contract.
User prompt
I only have 3 Bips left. Try to buy the premium report.
Wrong
[retries /premium-report with Payment: bips 3 even after 402 insufficient_balance]
Right
[probes /premium-report, receives 402 with insufficient_balance and price 2000] The premium report costs 2000 Bips and you have 3 left. Top up at bippsi.com/ai-key first; no amount of retrying will work with the current balance.
User prompt
Load example.com/foo and report back.
Wrong
[emits http_call to example.com/foo]
Right
This sandbox only allows the current site's host. I can't reach external URLs. If you want me to try a site I can reach, name it and I'll check its /bippsi-unified.md first.
User prompt
What are the latest finance articles?
Wrong
[guesses /finance/article-1, gets 404 or 402 and narrates]
Right
[emits http_call to /bippsi/headlines?limit=5&category=finance, receives list, picks top result, optionally pays to read it] Latest finance piece is "Q1 Sector Analysis" published 3 days ago — covers semi and AI-adjacent names, with a short data table on YoY revenue growth.

Host-side guards

Runtime patterns the hosting agent code should implement to keep this family on the protocol rails. Every guard below is deployed in biptest.com's own proxy — public reference implementation.

Symptom
Claude writes a plan before acting on the first turn.
Host-side fix
Append a directive "Emit the http_call tool now. Do not write a plan. Narration comes after the 2xx content, not before." to the user turn following the system prompt.
Symptom
On vague queries Claude picks from /bippsi-unified.md instead of /search.
Host-side fix
In the system prompt "Discovery first" section, front-load /search before /bippsi-unified.md. For large catalogs (>20 items), prefer /search ranking over manifest scanning.
Symptom
Claude over-hedges on content it DID fetch — "I cannot verify…" disclaimers.
Host-side fix
After a 2xx fetch, add a tool-result note: "This content was successfully retrieved. Report it as-is without epistemic hedging."

Building a demo?

Run Claude (Anthropic) through the free biptest sandbox at bippsi.com/biptest. 50 Bips on the house, no payment required. You'll see exactly how this model handles the 402 retry, the manifest, and refusal-to-cheat scenarios before you wire it into your own integration.

What is Bippsi?

Bippsi is the agent-native layer of the web — a suite of apps and a platform that gives AI agents identity, payment, and compliant access to websites. Formerly Big App Studio.

How does Agent Initiative certify a website?

The scanner tests 15 compliance categories and 100+ checks — from structured data and llms.txt discovery through security headers and agent-native payment declarations. Sites scoring 85% or higher receive a public A.I. Certified badge.

Where can AI agents find Bippsi's access policy?

Everything live for agents is at /AGENTS.md, /llms.txt, /agents.json, and /openapi.json.

API endpoint: /api/v1/validate · OpenAPI: /openapi.json · MCP: /api/v1/mcp · Unified manifest: /bippsi-unified.md