Your model charges to read everything sent to them. Needlepath hands it only what matters.

Copy, paste, run.

One key, one call before your model call, and a dashboard to watch it. The code below is the same as the docs; swap in your records and your prompt.

1. 1
   Sign up and get an API key

   Sign up for public beta today. The key is minted in the console and lives in your environment as NEEDLEPATH_API_KEY.

2. 2
   Send your records and your prompt, get the context back, call your model

   PythonTypeScriptLangChainLiteLLMShell

   pip install needlepathCopy

   import os
   from needlepath import NeedlepathClient, ContextRecord, TaskSpec
   
   client = NeedlepathClient(
       api_key=os.environ["NEEDLEPATH_API_KEY"],
       operating_point="np-2026-08-r4",
   )
   
   # What the agent holds for this question, one record each. For example:
   #   every purchase order from the last two years
   #   the supplier's contract and delivery terms
   #   the conversation so far
   records = [
       ContextRecord(text=po.text, kind="external_data", id=po.id, title=po.title)
       for po in purchase_orders
   ]
   prompt = "Which Corvane Freight purchase orders are still open past their delivery date?"
   
   result = client.select(records=records, task=TaskSpec(prompt=prompt), max_context_tokens=4000)
   context = result.rendered_context if result.applied else "\n\n".join(r.text for r in records)
   
   # Call your model as you already do, with the context ahead of the prompt.
   from openai import OpenAI
   answer = OpenAI().chat.completions.create(
       model="your-model",
       messages=[{"role": "user", "content": f"{context}\n\n{prompt}"}],
   )

3. 3
   Watch it work

   Requests, tokens metered and tokens saved, per key, in the console. When Needlepath stands aside, the whole set goes through unchanged.

Full clients, the LlamaIndex adapter and the rules every adapter follows: docs.nextmoca.com/integrations.