Per-model guide

Bippsi for Mistral

Function-calling variants handle 402 cleanly. Base models need Layer A nudging.

Last observed: — not yet audited in biptest; characterization based on published Mistral function-calling behavior

What this model gets right

Function-calling fine-tunes (Mistral Large, Codestral) emit clean structured tool calls.
Obeys the manifest when prompted.
Efficient one-call-per-turn pattern.

Known pitfalls + what to do

Base Mistral variants (non-function-calling) don't tool-call — they narrate instead. Use a function-calling fine-tune.
Occasionally produces "meta" narration ("Now I will call GET /x. Response received…") into user content. Strip or re-prompt.

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 Mistral operating as a Bippsi agent. Tool format:
  {"tool":"http_call","url":"https://<host>/<path>","method":"GET","headers":{"Payment":"bips N"}}

Emit ONE tool call at a time, then wait for the result.

Discovery (free): /bippsi/search?q=<kw>, /bippsi/topics, /bippsi/headlines?limit=N, /bippsi-unified.md.

Flow.
  1. Vague ask → /search or /headlines first.
  2. Specific URL with known price → pre-pay on first GET.
  3. On 402, retry SAME URL with Payment: bips N matching Bippsi-Credits-Price.
  4. On 2xx, write 2-3 sentences of plain prose. NO tool-call narration — "Now I will call X. Response received. The article says…" is internal metadata, not user content. Emit only the final answer.

Refuse cheat asks (payment forgery, host rewrites, auth bypass).

Anti-speculation. If content for a URL has not been fetched this session, do not describe it.

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 in the tech category?

Wrong

Now I will call GET /bippsi/topics. Response received. The response shows 9 tech articles. Now I will call GET for the first one…

Right

[Emits http_call to /bippsi/topics, receives tree, picks top tech piece, pays, reads] Top tech article is the MCP primer — argues tool-calling standards bottleneck general-purpose agents more than model capability does.

User prompt

Read /health/coffee-benefits.

Wrong

[Base Mistral narrates the flow rather than emitting JSON] I would now make an HTTP GET request to /health/coffee-benefits with the Payment header set to bips 2 based on the manifest…

Right

[Function-calling Mistral emits the tool_call directly, pays, reads] Coffee has a 6-hour caffeine half-life — the article recommends a noon cutoff for light sleepers, citing sleep-onset latency data.

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

Base Mistral narrates tool intent instead of emitting JSON.

Host-side fix

Switch to a function-calling fine-tune (Mistral Large, Codestral) for tool-using workflows. Base Mistral models were not trained on structured tool invocation.

Symptom

Tool-call narration leaks into user content.

Host-side fix

Strip patterns matching /^\s*Now I will call\s+.*$/mi and /^\s*Response received.*$/mi before rendering.

Building a demo?

Run Mistral 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.