Databases

Call databases are tables of structured data extracted after every attached agent’s call. They are not a generic SQL store and not the knowledge base.

Databases — one row per finished call (web, inbound, outbound).
Knowledge bases — files for mid-call RAG search.

Requires feature_databases. Console: Monitor → Databases, or attach a database on the agent editor.

How a row is written

Create a database with a name and columns (fields).
Attach one or more agents.
When a call hangs up, Kupe runs post-call extraction and inserts one row per session.

Always-present columns:

Column	Meaning
`who_called`	Caller phone or web participant
`direction`	`inbound` (incoming), `outbound`, or `web`
`started_at`	Call start
`duration_seconds`	Length

Schema you define, plus built-in analysis fields summary (string) and success (boolean). Custom fields:

{ "name": "promise_date", "type": "string", "description": "Date they promised to pay" }

type is string | number | boolean | enum (enum needs enum_values). Unique on (database_id, session_id).

Create and attach

from kupe import Kupe

client = Kupe()
db = client.databases.create(
    name="Collections outcomes",
    description="Promise dates from EMI calls",
    fields=[
        {"name": "promise_date", "type": "string", "description": "Date they promised to pay"},
        {"name": "amount", "type": "number", "description": "Amount confirmed"},
        {
            "name": "intent",
            "type": "enum",
            "description": "Will pay / dispute / DND",
            "enum_values": ["will_pay", "dispute", "dnd"],
        },
    ],
)
client.databases.attach_agent(db.id, agent_id="agt_...")
rows = client.databases.rows(db.id, q="pending")

import { Kupe } from "kupe-sdk";

const kupe = new Kupe();
const db = await kupe.databases.create({
  name: "Collections outcomes",
  description: "Promise dates from EMI calls",
  fields: [
    { name: "promise_date", type: "string", description: "Date they promised to pay" },
    { name: "amount", type: "number", description: "Amount confirmed" },
  ],
});
await kupe.databases.agents.attach(db.id, { agent_id: "agt_..." });
const rows = await kupe.databases.rows.list(db.id, { q: "pending" });

ORG=$(curl -s https://x.kupe.in/v1/me -H "Authorization: Bearer sk-kupe-YOUR_KEY" | jq -r .org_id)
PRJ=$(curl -s https://x.kupe.in/v1/me -H "Authorization: Bearer sk-kupe-YOUR_KEY" | jq -r .project_id)

curl -X POST "https://x.kupe.in/v1/orgs/${ORG}/projects/${PRJ}/databases" \
  -H "Authorization: Bearer sk-kupe-YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Collections outcomes",
    "fields": [
      {"name": "promise_date", "type": "string", "description": "Date they promised to pay"}
    ]
  }'

You can also pass agent_ids on create to attach in the same request.

Routes

Action	Path
List / create	`GET\|POST /v1/orgs/{org_id}/projects/{project_id}/databases`
Get / patch / archive	`/v1/databases/{database_id}`
Agents	`GET\|POST /v1/databases/{id}/agents`, `DELETE .../agents/{agent_id}`
Reverse	`GET /v1/agents/{agent_id}/databases`
Rows (keyset)	`GET /v1/databases/{id}/rows?cursor=&limit=&q=`
Patch / delete a row	`PATCH\|DELETE /v1/databases/{id}/rows/{row_id}`
Export	`GET /v1/databases/{id}/export?format=csv\|json\|ndjson\|zip`

Patch fields to replace the extraction schema for every attached agent.

Destinations

After a row is written you can fan it out. Patch destinations:

[
  { "kind": "webhook", "url": "https://example.com/hooks/calls" },
  { "kind": "tool", "tool_id": "tool_..." },
  { "kind": "composio", "composio_slug": "hubspot", "connected_account_id": "ca_..." }
]

Rows

Each row includes direction (web | inbound | outbound), values (JSON of schema fields), status (pending | skipped | completed | failed), and sync_error. Search q matches caller, direction, and values text. The console sorts, filters empty columns, hides columns, and exports. Right-click a row to edit values, change a column type, or delete it.

blob = client.databases.export(db.id, format="csv")
open("calls.csv", "wb").write(blob)

const buf = await kupe.databases.rows.export(db.id, { format: "csv" });

curl "https://x.kupe.in/v1/databases/db_.../export?format=csv" \
  -H "Authorization: Bearer sk-kupe-YOUR_KEY" -o calls.csv

Knowledge bases

Separate product for RAG files. Attach to an agent with config.knowledge_base_ids — do not dump FAQs into the system prompt.

kb = client.knowledge_bases.create(name="Policy")
client.knowledge_bases.files.upload(kb.id, file=open("policy.pdf", "rb"))
hits = client.knowledge_bases.search(kb.id, query="prepayment charges", top_k=5)

const kb = await kupe.knowledgeBases.create({ name: "Policy" });
await kupe.knowledgeBases.files.upload(kb.id, file);
const hits = await kupe.knowledgeBases.search(kb.id, {
  query: "prepayment charges",
  top_k: 5,
});

curl -X POST "https://x.kupe.in/v1/orgs/${ORG}/projects/${PRJ}/knowledge-bases" \
  -H "Authorization: Bearer sk-kupe-YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"name":"Policy"}'

Audio assets

Ambient beds for web calls: GET|POST /v1/orgs/{org_id}/projects/{project_id}/audio-assets, archive with POST /v1/audio-assets/{asset_id}/archive.

Get started

Realtime

Platform

MCP

How a row is written

Create and attach

Routes

Destinations

Rows

Knowledge bases

Audio assets

​How a row is written

​Create and attach

​Routes

​Destinations

​Rows

​Knowledge bases

​Audio assets

How a row is written

Create and attach

Routes

Destinations

Rows

Knowledge bases

Audio assets