We make your AI governance blazing fast.

from langchain.agents import create_agent
from needlepath_langchain import NeedlepathMiddleware

# NEEDLEPATH_API_KEY is read from the environment. Nothing else in the agent changes.
agent = create_agent(model, tools, middleware=[NeedlepathMiddleware(operating_point="np-2026-08-r4")])

answer = agent.invoke({"messages": [("user", "Which Corvane Freight purchase orders are still open past their delivery date?")]})

import os
from needlepath import NeedlepathClient, ContextRecord, TaskSpec
from openai import OpenAI

client = NeedlepathClient(api_key=os.environ["NEEDLEPATH_API_KEY"], operating_point="np-2026-08-r4")

# One record per item the agent holds: purchase orders, the supplier contract, 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)

# Your model call, unchanged, with the context ahead of the prompt.
answer = OpenAI().chat.completions.create(
    model="your-model", messages=[{"role": "user", "content": f"{context}\n\n{prompt}"}]
)

import { NeedlepathClient } from "@nextmoca/needlepath-sdk";
import OpenAI from "openai";

const client = new NeedlepathClient({ apiKey: process.env.NEEDLEPATH_API_KEY!, operatingPoint: "np-2026-08-r4" });

// One record per item the agent holds: purchase orders, the supplier contract, the conversation so far.
const records = purchaseOrders.map((po) => ({ text: po.text, kind: "external_data", id: po.id, title: po.title }));
const prompt = "Which Corvane Freight purchase orders are still open past their delivery date?";

const result = await client.select({ records, task: { prompt }, maxContextTokens: 4000 });
const context = result.applied ? result.response!.renderedContext : records.map((r) => r.text).join("\n\n");

// Your model call, unchanged, with the context ahead of the prompt.
const answer = await new OpenAI().chat.completions.create({
  model: "your-model", messages: [{ role: "user", content: `${context}\n\n${prompt}` }],
});

# config.yaml: every client behind the proxy gets selection, with no client-side change.
guardrails:
  - guardrail_name: "needlepath"
    litellm_params:
      guardrail: needlepath_litellm.NeedlepathGuardrail
      mode: "pre_call"
      default_on: true
      operating_point: "np-2026-08-r4"
      history_max_tokens: 8000
      preserve_recent: 2

# export NEEDLEPATH_API_KEY="np_live_..." in the proxy's environment, then: litellm --config config.yaml

# One record per item the agent holds. Use rendered_context from the response ahead of the prompt.
curl https://api.nextmoca.com/v1/context/select \
  -H "Authorization: Bearer $NEEDLEPATH_API_KEY" -H "Content-Type: application/json" \
  -d '{
    "records": [
      { "id": "po-2026-0412", "kind": "external_data", "title": "PO 2026-0412, Corvane Freight", "text": "..." },
      { "id": "contract-corvane", "kind": "external_data", "title": "Corvane Freight delivery terms", "text": "..." }
    ],
    "task": { "prompt": "Which Corvane Freight purchase orders are still open past their delivery date?" },
    "budget": { "max_context_tokens": 4000, "operating_point": "np-2026-08-r4" },
    "render": true
  }'