GOPURAM

Summarization

Any length in, summary out — chunking and map-reduce handled server-side, every step itemized.

POST /v1/summarize summarizes anything — a paragraph, a 400-page PDF, a long meeting transcript — in one call. When the input exceeds the model's context window, the gateway chunks it on paragraph boundaries, summarizes the pieces, and merges them (recursively if needed) — the map-reduce plumbing every team hand-rolls, done server-side and itemized on the bill.

curl https://api.gopuram.net/v1/summarize \
  -H "Authorization: Bearer $GOPURAM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "input": {"type": "document_url", "document_url": "data:application/pdf;base64,<PDF>"},
    "length": "brief",
    "style": "bullets"
  }'

{
  "object": "summary",
  "model": "google/gemini-3.6-flash",
  "summary": "- The contract term is 24 months…",
  "usage": {
    "cost": 0.0031,
    "input_tokens": 41200,
    "output_tokens": 950,
    "steps": [
      { "kind": "ocr", "model": "mistral/mistral-ocr", "pages": 38, "cost": 0.1596 },
      { "kind": "llm", "stage": "map", "chunk": 0, "cost": 0.0009 },
      { "kind": "llm", "stage": "map", "chunk": 1, "cost": 0.0008 },
      { "kind": "llm", "stage": "reduce", "cost": 0.0002 }
    ]
  }
}

Request

FieldTypeNotes
inputobject{type:"text", text} (≤8M chars) · {type:"document_url", …} (PDF via OCR) · {type:"image_url", …}
length"brief" | "standard" | "detailed"≈100 / 300 / 800 words (default standard)
max_wordsint 1–5000overrides length (send one or the other)
style"prose" | "bullets" | "tldr"default prose
modelstringoptional chat-model override

Short inputs are a single model call (one steps entry). Long inputs show the whole pipeline — map steps per chunk, then reduce — each priced at the model's normal rate, exact total in usage.cost. Inputs beyond 64 chunks per pass are rejected with a clear 400 rather than silently truncated.