● Karaya is in private beta

An AI money assistant for the whole Indonesian household.

EkoNugroho Labs builds Karaya: a shared family ledger plus an AI assistant, designed around Claude's tool use, that records spending from a chat message and answers questions like “how much did we spend on school this month?” using the family's real numbers.

Request beta access See how it works →
Market
Indonesian families
AI
Claude tool use
Stage
Bootstrapped
K

Karaya Assistant

Rumah Tangga Nugroho · 2 members

example
Tadi bayar SPP Rp850rb pakai rekening bersama, sama jajan anak 25rb cash
→ create_transaction(850000, "Pendidikan", account="Rekening Bersama")
→ create_transaction(25000, "Jajan Anak", account="Cash")
Sudah dicatat ✅ 2 transaksi. Pengeluaran pendidikan bulan ini Rp850.000 (anggaran Rp1.000.000, sisa Rp150.000).
Bulan ini kita boros di mana?
Kategori Makan di luar naik 42% dibanding bulan lalu, sebagian besar dari akun pribadi Ayah. Mau saya buatkan batas mingguan?
Illustrative example. Names and amounts are not real user data.

The problem

Family money in Indonesia is shared, but the tools are not.

A typical household runs on a mix of joint and personal bank accounts, e-wallets and cash, managed by two or more people. Budgeting apps assume one user with one wallet, and spreadsheets break the moment nobody updates them. The result: couples argue about numbers nobody actually has.

01 · Shared household

One ledger, many members

Husband, wife and optional viewers share one household. Accounts can be personal or joint, and reports can be combined, per member, or per account.

02 · Chat-first input

Record money the way people talk

“Bayar listrik 450rb” is enough. The assistant resolves the category and account by name, asks when something is ambiguous, and never invents a transaction.

03 · Answers from real data

Grounded, not guessed

Every answer about money comes from a tool call against the household's own ledger: balances, budgets, trends and monthly summaries in plain Indonesian.

How it works

Claude does the reasoning. Karaya's API does the bookkeeping.

We don't let a language model “remember” anyone's finances. Claude receives a typed set of Karaya tools and decides which to call; the numbers always come back from our API. This keeps answers auditable and makes the same tools usable by any agent. During the beta, households connect their own AI key (Claude, or another provider); our next step is to make Claude the default, managed model so no family needs a key of its own.

  • 1

    Tool use over the household ledger

    The assistant is instructed to always fetch data through tools before answering, so summaries, budgets and trends are computed, not hallucinated.

  • 2

    Agent-first public API (/v1)

    REST resources described in OpenAPI 3.1, long-lived scoped API keys, idempotency keys, tolerant name resolution and actionable errors, designed for agents rather than only for our own frontend.

  • 3

    Chat where families already are

    Today: the web app and a Telegram agent connected through the /v1 API. Next: an MCP server so Claude Desktop and Claude Code users can work with their household data directly.

architecture 
# one request, end to end
user (web · Telegram)
   │  "bulan ini kita boros di mana?"
   ▼
Claude  ── tool_use ──▶  get_spending_trend(month)
   ▲                         │
   │                         ▼
   │                 Karaya API /v1  (Hono · TypeScript)
   │                         │  household-scoped auth
   │                         ▼
   │                 Postgres (Supabase)
   │                         │
   └──── tool_result ◀───────┘
   ▼
answer in Indonesian, grounded in the ledger

Backend

TypeScript · Hono · Supabase Postgres

Data model

Household-scoped, multi-member

API contract

OpenAPI 3.1 · idempotent writes

Deploy

Vercel · Supabase Edge Functions

Roadmap

Where Karaya is going

  1. done

    Household platform

    Multi-member households, personal and joint accounts, the agent-first /v1 API, and the tool-using AI assistant.

  2. now

    Private beta

    Onboarding a small group of families, tuning the assistant's Indonesian prompts and tools on real household questions.

  3. next

    Claude by default, MCP & family features

    Claude as the built-in model for every household, an MCP server for Claude Desktop and Claude Code, automatic monthly reports in chat, bank statement reconciliation, and household shopping and stock.

Also from the lab

Smaller products we ship and run

Each one is built, deployed and operated in-house. They share the same idea: useful agents need a clean, typed API underneath.

Cortex AI

live

Managed, isolated AI agent workspaces you control from your own Telegram bot. No terminal needed. One-minute setup, daily backups, pay per day.

cortex-ai.my.id →

Karaya Marketing

in development

Content planning, AI copywriting and scheduling for Indonesian digital-product creators, with a REST API and an MCP adapter so agents such as Claude Desktop can operate it.

Not yet public

Pocket Neon Arcade

live

A collection of ten neon 3D arcade mini-games for Android, published on Google Play. Our first shipped consumer app.

Google Play →
Eko Nugroho, founder

Founder

Eko Nugroho

Backend engineer with 7+ years building high-load systems in Go and Java: transaction APIs, data migrations and microservice re-architectures across transportation, travel and banking. EkoNugroho Labs is where that experience goes into products of our own, starting with the money problem every Indonesian family has.

LinkedIn GitHub Company page

Private beta

Want Karaya for your household?

Tell us a little about your family's setup (how many people, which accounts) and we'll get you onboarded. Partners and investors are welcome to reach out too.

founder@ekonugroho.dev

Bogor, Indonesia