Docs & API

Build with P34

One question, one API: given the trade options in front of you right now, which should you take, at what size, and what profit should you expect? Everything here is built to be driven by your agent — or by you.

Platform locations

Where everything lives

Quickstart

Fit → poll → portfolio

  1. Register at the console and subscribe to a plan.
  2. Create an API key — test-… for integration, profit-… for production decisions.
  3. Prepare two tables and a config: Menus (every option you faced — including the deals you declined), Sales (your realized outcomes log), market_type (your market's economics).
  4. POST /fit — returns a session_id in seconds.
  5. Poll GET /result/{session_id} until done — or watch it live in the console.
  6. Receive your T=0 menu filled in: quantity per key (qty = 0 = do not trade) and predicted profit, calibrated as a portfolio sum.

Tables travel as JSON records or base64 Parquet. An optional confidence_correction in [−1, 1] tilts selection: positive = fewer, higher-confidence picks; negative = more scenarios.

# pip install pandas requests pyarrow
import requests

API = "https://api.hyperc.com/v1"
H = {"Authorization": "Bearer $P34_API_KEY"}

r = requests.post(f"{API}/fit", headers=H, json={
    "menus": menus,          # history + the T=0 task menu
    "sales": sales,          # realized sales event tape
    "market_type": config,   # holding costs, write-offs, fees
    "model": "rc012"})
sid = r.json()["session_id"]

res = requests.get(f"{API}/result/{sid}", headers=H).json()
# status: queued → processing → done | failed
# done ⇒ selected qty + predicted profit per key,
#        n_selected, predicted_profit_sum
Reference

Endpoints

Base URL: https://api.hyperc.com/v1 · Auth: Authorization: Bearer <key>

EndpointWhat it does
GET /Service info: protocol, available model versions, endpoint list. Open.
GET /healthLiveness probe. Open.
POST /fitSubmit Menus + Sales + market_type. Validates, grounds the history against your market economics, enqueues the calculation on the compute cluster. Returns session_id immediately with intake statistics.
GET /result/{session_id}Poll the calculation: queued → processing → done / failed. done carries the predicted T=0 menu with selected quantities and profit, plus confidence fields.
POST /predictInstant selection from a small in-process reference model — a payload sanity-checker while the real calculation runs. Not P34's answer; /result is.
Validation with teeth

The task menu must carry no outcome values — P34 never sees your future. And the service refuses histories that are all wins: declined-deal context is a mandatory training input, not a nicety.

Model versions

GET / lists offered versions (e.g. r008, rc012, rc012-ray). The chosen version runs the whole calculation and is echoed in the response.

Limits

Compute budgets scale with your plan and show as percentage used. Weekly and rolling five-hour windows apply; request size caps at 300 MB.

Documentation

The docs repo, mapped

Everything is public at github.com/hyperc-ai/P34-API-DOCS — docs, runnable examples, sample data and a CI-ready test workflow.

PathContents
docs/01-overview.mdWhat P34 does and the mental model: menus, sales, the T=0 task, why not just a regressor.
docs/02-endpoints.mdEndpoint reference, auth, result statuses, model versions, confidence correction, wire formats, limits.
docs/03-data-format.mdThe Menus / Sales / market_type input format, rule by rule — including “include the deals you did not take.”
docs/04-errors-and-checks.mdCommon validation errors and quick self-checks.
examples/client/A complete sample client: fit → poll → portfolio, with JSON/Parquet wire helpers.
examples/data/Sample input as Excel, CSV and JSON.
examples/pytest/A minimal pytest workflow to validate your integration in CI.
examples/baseline_comparison/P34 vs a gradient-boosting profit regressor on a synthetic market with known ground truth.
Open source

github.com/hyperc-ai

The public P34 repositories. Simulator and Arena repositories are coming soon.

P34-API-DOCS

Complete user-facing P34 API documentation: endpoint reference, data format, examples, pytest workflow, baseline comparison.

Markdown + Python · active
p34-technical-report

P34 Technical Report and results discussion — benchmark notebooks behind “Learning When Not to Trade.”

Jupyter Notebook · ★ 31 · active
simulator · arena

The market-simulator repository and the PARML Arena are being prepared for publication.

coming soon

API Terms of Use

Use of the API, console and simulator is governed by the P34 API Terms of Use (effective 2026-07-20), presented at registration in the console. They cover tiers and tokens, profit-share pricing, regulated-market classification, acceptable use, data handling and liability.

Read the API Terms in the console →

Membership Terms of Use (this site) →

The Membership Terms govern the membership program (community, opportunity feed, founding economics); the API Terms govern the API service itself. Where you use both, both apply.