Feature Description

Please raise or remove the 20-question-per-call cap for typesafe/jev on POST /provider/v1/systemone, so larger sets of independent questions can share one state and one request.

Ideally, the endpoint would accept question maps within the upstream model's documented token/request budgets. If an operational question-count cap is necessary, please consider a higher limit and document its rationale and supported maximum.

Use Case

My application already splits larger question sets into batches of 20 to use Command Code's Jev endpoint. Every batch repeats the same state, adding HTTP requests and repeated input processing/billing. A higher cap would let applications use Jev's intended one-state/many-questions pattern more efficiently.

TypeSafe explicitly recommends putting the questions a system needs into a single request and evaluating them in parallel: Speculative fan-out.

Additional Context

Direct HTTP tests performed on 2026-10-07, using Windows Python and httpx, with a short synthetic state and identical Noul questions:

The Command Code response at 21 questions contains:

{
  "error": {
    "type": "invalid_request_error",
    "message": "at most 20 questions per call",
    "param": "questions"
  }
}

Minimal reproduction (set your keys locally; no real application data is required):

import os
import httpx

state = "This is a synthetic API validation test. The color is blue."

cases = [
    ("https://api.commandcode.ai/provider/v1/systemone",
     "typesafe/jev", "CMD_API_KEY", 20),
    ("https://api.commandcode.ai/provider/v1/systemone",
     "typesafe/jev", "CMD_API_KEY", 21),
    ("https://api.typesafe.ai/v1/systemone",
     "jev-latest", "TYPESAFE_API_KEY", 21),
]

for url, model, key_env, count in cases:
    questions = {
        f"q{i:02}": {"type": "noul", "instructions": "Is the color blue?"}
        for i in range(count)
    }
    response = httpx.post(
        url,
        headers={"Authorization": "Bearer " + os.environ[key_env]},
        json={"model": model, "state": state, "questions": questions},
        timeout=15,
        follow_redirects=False,
    )
    data = response.json()
    print(model, count, response.status_code,
          len(data.get("answers", {})), data.get("error"))

The Command Code Provider API documentation describes the TypeSafe request/response shape, but I could not find the 20-question cap in that page or the Jev model page.

TypeSafe's model documentation describes a 64K-token budget for the state plus all questions, and a 32K-token budget for the state plus the longest question. The 21-question control test above establishes that the observed 20-question cap is specific to the Command Code serving path; it does not identify which internal layer applies it.

Could you clarify why the current cap is 20 and whether it can be raised or removed?

How important is this to you?

Important for my workflow