API & CLI

Endpoint

multipart/form-data. Authenticate with Authorization: Bearer fp_live_.... The response streams back as raw text/plain, no JSON envelope, no SSE framing, read the bytes as they arrive and concatenate.

Fields

curl

curl -N -X POST https://forgepilot-production-9575.up.railway.app/analyze \
  -H "Authorization: Bearer fp_live_your_key_here" \
  -F "prompt=FMEA for a bolted flange assembly transmitting 5 kN axial load, sealing 200 bar hydraulic fluid, -40C to 120C operating range" \
  -F "files=@bracket_drawing.pdf"

Python

import requests

API_KEY = "fp_live_your_key_here"
BACKEND = "https://forgepilot-production-9575.up.railway.app"

with requests.post(
    f"{BACKEND}/analyze",
    headers={"Authorization": f"Bearer {API_KEY}"},
    data={"prompt": "Tolerance stack-up for a 4-part shaft/bearing/housing chain, worst-case and RSS"},
    files={"files": open("assembly.step", "rb")},
    stream=True,
) as r:
    r.raise_for_status()
    for chunk in r.iter_content(chunk_size=None, decode_unicode=True):
        print(chunk, end="", flush=True)

To keep a multi-step session coherent (ask a follow-up that references a result you already got), pass the prior turns back as history:

# Follow-up turn in the same session: pass back what already happened
# as {role, content} pairs so the model has real context, not a cold prompt.
import json

history = [
    {"role": "user", "content": "[Ran Drawing Review (files: bracket_drawing.pdf)]"},
    {"role": "assistant", "content": "<the first response's text>"},
]

requests.post(
    f"{BACKEND}/analyze",
    headers={"Authorization": f"Bearer {API_KEY}"},
    data={
        "prompt": "What about the tolerance on the mounting hole we flagged?",
        "history": json.dumps(history),
    },
    stream=True,
)

Rate limits

Shared with your web workspace usage: an API call and a UI click draw from the same daily count.

Errors