REST API.

Authorization: Bearer <YOUR_CONTEXT_API_KEY>

# 1. Record an exchange (send full text, verbatim)
curl -X POST https://api.belleq.app/v1/record \
  -H "Authorization: Bearer $BELLEQ_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "user_message": "We decided to launch on June 30.",
    "assistant_message": "Got it — launch date set to June 30.",
    "conversation_id": "planning-2026"
  }'

# 2. Index it right away
curl -X POST https://api.belleq.app/v1/flush \
  -H "Authorization: Bearer $BELLEQ_KEY"

# 3. Search for it later
curl -X POST https://api.belleq.app/v1/query \
  -H "Authorization: Bearer $BELLEQ_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query": "when are we launching?"}'

import os, requests
from google import genai
from google.genai import types

BELLEQ = "https://api.belleq.app/v1"
KEY = os.environ["BELLEQ_KEY"]

def belleq(op: str, **body):
    r = requests.post(f"{BELLEQ}/{op}",
        headers={"Authorization": f"Bearer {KEY}"}, json=body)
    r.raise_for_status()
    return r.json()

recall = types.FunctionDeclaration(
    name="recall", description="Load recent Belleq memory at the start of a chat.",
    parameters={"type": "object", "properties": {"limit": {"type": "integer"}}})
query = types.FunctionDeclaration(
    name="query", description="Search Belleq memory for a topic.",
    parameters={"type": "object", "properties": {"query": {"type": "string"}},
                "required": ["query"]})

client = genai.Client()
tools = types.Tool(function_declarations=[recall, query])
# When Gemini calls a function, dispatch it: belleq(call.name, **call.args)

import os, requests

BELLEQ = "https://api.belleq.app/v1"
H = {"Authorization": f"Bearer {os.environ['BELLEQ_KEY']}"}

# Search
hits = requests.post(f"{BELLEQ}/query", headers=H,
    json={"query": "launch date"}).json()
for c in hits["chunks"]:
    print(round(c["score"], 2), c["text"])

# Save + index
requests.post(f"{BELLEQ}/record", headers=H, json={
    "user_message": "...", "assistant_message": "...",
    "conversation_id": "session-1"})
requests.post(f"{BELLEQ}/flush", headers=H)

You have a persistent memory tool, Belleq. Use it automatically:
1. At the START of a conversation, call recall to load prior context.
2. When a question needs past detail, call query before answering.
3. After EACH reply, call record with the user's message and your reply
   verbatim (do not summarize). Reuse one conversation_id for the chat.
4. Call flush when the conversation winds down.
Do this silently, without asking permission.