Building context

Before each model call, your app asks EDN for the memory relevant to the user’s message. This guide shows the common cases. For what each response field means, see The SIP. The examples use the EDN_API_KEY and EDN_SESSION_ID environment variables from the Quickstart, and this setup:

import os
import requests

EDN = "https://api.tensorpro.ai/api/v1"
HEADERS = {"Authorization": f"Bearer {os.environ['EDN_API_KEY']}"}
session_id = os.environ["EDN_SESSION_ID"]
user_text = "Who manages the website project?"

const EDN = "https://api.tensorpro.ai/api/v1";
const HEADERS = {
  Authorization: `Bearer ${process.env.EDN_API_KEY}`,
  "Content-Type": "application/json",
};
const sessionId = process.env.EDN_SESSION_ID;
const userText = "Who manages the website project?";

Get context for a turn

Build a SIP with the user’s latest message, then put sip in your model’s system prompt:

curl https://api.tensorpro.ai/api/v1/sip/build \
  -H "Authorization: Bearer $EDN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query": "Who manages the website project?", "session_id": "'"$EDN_SESSION_ID"'"}'

sip = requests.post(
    f"{EDN}/sip/build",
    headers=HEADERS,
    json={"query": user_text, "session_id": session_id},
).json()

system_prompt = "You are a helpful assistant."
if sip["sip"]:
    system_prompt += "\n\nUse this memory about the user when it is relevant:\n\n" + sip["sip"]

const sip = await (
  await fetch(`${EDN}/sip/build`, {
    method: "POST",
    headers: HEADERS,
    body: JSON.stringify({ query: userText, session_id: sessionId }),
  })
).json();

let systemPrompt = "You are a helpful assistant.";
if (sip.sip) {
  systemPrompt += `\n\nUse this memory about the user when it is relevant:\n\n${sip.sip}`;
}

If sip is empty, there is no relevant memory, so call your model without it.

Get context from a project

To answer from one project’s memory and documents, add project_id. Add include_global: true if general memory should be included too:

curl https://api.tensorpro.ai/api/v1/sip/build \
  -H "Authorization: Bearer $EDN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "Who manages the website project?",
    "session_id": "'"$EDN_SESSION_ID"'",
    "project_id": "YOUR_PROJECT_ID",
    "include_global": false
  }'

sip = requests.post(
    f"{EDN}/sip/build",
    headers=HEADERS,
    json={
        "query": user_text,
        "session_id": session_id,
        "project_id": project_id,
        "include_global": False,
    },
).json()

const sip = await (
  await fetch(`${EDN}/sip/build`, {
    method: "POST",
    headers: HEADERS,
    body: JSON.stringify({
      query: userText,
      session_id: sessionId,
      project_id: projectId,
      include_global: false,
    }),
  })
).json();

Project builds number their sources, so ask your model to cite them:

system_prompt = (
    "Answer using only the sources below. Cite each source you use with its "
    "number, for example [1].\n\n" + sip["sip"]
)

const systemPrompt =
  "Answer using only the sources below. Cite each source you use with its " +
  `number, for example [1].\n\n${sip.sip}`;

Use citations to show each number’s source to your user. See Citations.

Help with follow-up questions

Short follow-ups like “and who approves it?” have little to search on. Save each message to the session as the conversation goes:

curl https://api.tensorpro.ai/api/v1/sessions/$EDN_SESSION_ID/messages \
  -H "Authorization: Bearer $EDN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"role": "user", "content": "Who manages the website project?"}'

requests.post(
    f"{EDN}/sessions/{session_id}/messages",
    headers=HEADERS,
    json={"role": "user", "content": user_text},
)

await fetch(`${EDN}/sessions/${sessionId}/messages`, {
  method: "POST",
  headers: HEADERS,
  body: JSON.stringify({ role: "user", content: userText }),
});

If a question on its own finds no relevant facts, EDN retries once using the user’s recent saved messages.

Search memory directly

To look up memory without building a SIP, for example for a search box in your app, use GET /api/v1/retrieval/search:

curl "https://api.tensorpro.ai/api/v1/retrieval/search?q=website%20project&limit=5" \
  -H "Authorization: Bearer $EDN_API_KEY"

results = requests.get(
    f"{EDN}/retrieval/search",
    headers=HEADERS,
    params={"q": "website project", "limit": 5},
).json()

const results = await (
  await fetch(`${EDN}/retrieval/search?q=${encodeURIComponent("website project")}&limit=5`, {
    headers: HEADERS,
  })
).json();

Response (shortened)

{
  "records": [
    {
      "id": "ef771387-4412-4784-a94c-73c819991fad",
      "content_summary": "Sara manages the website project.",
      "provenance": "USER_STATED",
      "domain": "factual",
      "similarity": 0.70,
      "created_at": "2026-11-10T06:36:10+00:00"
    }
  ],
  "count": 1
}

Results are ordered by similarity; higher means a closer match.
limit defaults to 10. Values above 50 are treated as 50.
Search covers general memory only, not project memory, and only records that can be used as context.

Next steps

The SIP

Every response field explained.

Writing memory

Save each turn after the model replies.

Quickstart

The full loop in five minutes.

Get started

Core concepts

Guides

Reference guides