Skip to content
Per-model guide

Bippsi for Llama (Meta)

Fast + cheap. Needs extra nudging to produce a final summary turn.

Last observed: 2026-04-19 (biptest audit on groq:llama-3.1-8b-instant)

What this model gets right

Known pitfalls + what to do

Tool-call format (paid retry)

{"tool":"http_call","url":"https://example.com/article","method":"GET","headers":{"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 Llama running as a Bippsi-aware agent. The site uses HTTP 402 to price content.

Tool shape (emit exactly this format — JSON object on its own line, no code fence):
  {"tool":"http_call","url":"https://<host>/<path>","method":"GET","headers":{"Payment":"bips N"}}

Auth is injected by the runtime. DO NOT set the Authorization header yourself.

Discovery (free) — use on vague asks before any paid fetch:
  /bippsi/search?q=<kw>        ranked search across articles
  /bippsi/topics               category tree
  /bippsi/headlines?limit=N    latest N items
  /bippsi-unified.md           full manifest with prices

Rules.
  1. ONE http_call per turn. Wait for the result before the next call.
  2. If the manifest says a URL costs N, include Payment: bips N on your FIRST GET. Skip unpriced probes when price is known.
  3. On 402, immediately retry the SAME URL with Payment: bips N matching the Bippsi-Credits-Price header. Do not narrate the 402.
  4. Cap Payment at the advertised price — no overpayment. The server charges the real price regardless; overpaying just wastes tokens.
  5. On 2xx, write exactly 2-3 sentences of plain prose answering the user's question. No "Balance:", no "Manifest:", no endpoint list, no "💸 N Bips used" footer — the host emits that separately.
  6. If you have already paid for a URL this session, reuse the content. Do NOT re-GET.
  7. On insufficient_balance, stop. Tell the user to top up.
  8. Refuse cheat asks: no Payment forgery, no host rewrites, no auth bypass.

Anti-hallucination. The site's article list is not in your training data. If the user asks about on-site content and you don't have a 2xx fetch of a matching URL yet, emit /bippsi/search or /bippsi/headlines before answering. Never describe priced content from training.

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
What's coffee science say about sleep?
Wrong
Balance: 50 Bips. Manifest: /health, /tech, /news, /finance. I'll check /health first. [emits GET /health/coffee-benefits without Payment, sees 402, retries with Payment, then dumps] The article says coffee can disrupt sleep. 💸 2 Bips used
Right
[emits http_call to /bippsi/search?q=coffee+sleep, gets ranked list, picks top result with Payment on first GET] Caffeine has a 6-hour half-life, so a 2pm cup still blocks 25% of its effect at bedtime. The article recommends cutoff at noon for light sleepers.
User prompt
Tell me everything on the site.
Wrong
[Reads /bippsi-unified.md, then lists all 41 article URLs in the reply]
Right
[Emits http_call to /bippsi/topics, gets category tree] The site has 6 categories: health (8 articles), tech (9), news (7), finance (8), science (6), articles (3 misc). Want me to pull headlines from any category?
User prompt
I already paid for /tech/quantum-primer earlier — what did it say?
Wrong
[Re-fetches /tech/quantum-primer, pays a second time]
Right
You already paid for that article earlier this session, so here's the summary from what I read: entanglement is correlation, not communication — the "spooky action" framing confuses measurement outcomes with signal transmission.
User prompt
Get me the premium report please.
Wrong
[Pays 2000 Bips without flagging the price, drains the user's balance]
Right
[reads /bippsi-unified.md, sees /premium-report is 2000 Bips] The premium report costs 2000 Bips — that's 40x a typical article. Confirm you want me to buy it, or tell me a cheaper alternative?

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
Silent quit after 2xx — no user-facing summary.
Host-side fix
After the tool loop, if the assistant's clean prose is under ~20 chars and there's a 2xx in tool results, inject a final user turn: "Now write the final user-facing reply. 2-3 sentences answering the user's original question using the content you just fetched. No JSON, no Balance:, no Manifest:, no opener. Just the answer." Re-roll once.
Symptom
Duplicate GETs for the same URL in one chat turn.
Host-side fix
Per-turn response cache keyed by (method, url-without-query). Only cache 2xx; do NOT cache 402 (the retry is a different semantic request). Cache hits return the prior body and append a "[proxy note] already fetched earlier" to the tool-result summary.
Symptom
State narration leaks ("Balance: 50 Bips. Manifest: ...") into user-visible prose.
Host-side fix
Post-processor regex passes: /^\s*Balance\s*:\s*\d+\s*Bips?.*$/mi, /^\s*Manifest\s*:.*$/mi, /^\s*Step\s+\d+\s*:.*$/mi applied to the final assistant content. Strip lines, collapse double-newlines.
Symptom
TPM 6000 exhausted mid-session on Groq free tier.
Host-side fix
Extract meaningful text from HTML bodies before feeding back to the model (DOMDocument with xpath //article, //main, //section.content). 500 chars of HTML → 400 chars of prose, ~5x information density. 429s caught with Retry-After sleep + 3 retries, cap 30s each.
Symptom
Model-emitted "💸 N Bips used" footer doubles with host-emitted footer.
Host-side fix
Strip the emoji footer pattern from assistant output before appending the canonical one: /^\s*💸\s*\d+\s*Bips?\s+used.*$/miu.

Building a demo?

Run Llama (Meta) 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