The same idea as AgentBuff, in the open: software an AI agent can operate safely, with MCP servers, scoped tokens and deterministic engines wherever the numbers matter.
The idea behind all four is the one behind AgentBuff: software that an AI agent can operate safely. Where numbers matter, a tested engine does the maths and the language model only writes the words; where an agent can change data, it goes through scoped tokens and, for money, a human confirmation.
AgentBuff Co-Founder
A web app in Bahasa Indonesia that walks a small-business founder from a raw idea to sourced market research, a business plan with a financial model, a brand identity, and an investor-ready proposal and pitch deck. It runs on the user's own Gemini or OpenAI key.
- Numbers from a deterministic engine: cost of goods, break-even, ROI, NPV and IRR are calculated by tested code; the model only proposes assumptions and writes the narrative.
- Callable by agents: an MCP gateway with 9 tools and read or write scoped tokens.
- Stack: Next.js 16, React 19, strict TypeScript, Prisma with PostgreSQL, Vitest; the README lists 260 tests.
FinFreedom
A personal-finance app made for Indonesia: accounts and transactions, CSV import with automatic categories, four budgeting methods, and income tax (PPh 21), zakat and BPJS calculations.
- FIRE planning with 1,000 Monte Carlo simulations, backtests and stress tests.
- An MCP server with 13 tools, so an agent such as Claude or Cursor can read and record data, with scoped tokens, an audit log and a human confirmation before anything is written.
- Stack: a TypeScript monorepo with Next.js 15, React 19, Prisma and the official MCP SDK.
Linky
An open alternative to Bit.ly: custom short links, branded QR codes, click analytics, password-protected and expiring links, deep links for iOS and Android, link-in-bio pages, and CSV import from Bit.ly, Rebrandly, TinyURL, Dub.co and Short.io.
- For developers: a REST API with API keys, webhooks signed with HMAC-SHA256, and an OpenAPI 3.1 specification.
- Checked on every change: 146 test cases and a CI pipeline that runs lint, type checks, tests, the build, npm audit and Gitleaks.
- Stack: Next.js 15, React 19, strict TypeScript and Drizzle ORM on SQLite or PostgreSQL.
AgentBuff Presentation Skills
A skill for AI coding agents such as Claude Code, Codex and Gemini CLI: the agent builds an HTML presentation from a brief or a PowerPoint file and exports it to PDF, PNG or JPG, or PPTX that looks exactly like the original.
- Built on open source: it extends the Frontend Slides skill and the Bold Template Pack by zarazhangrui, credited in the README.
- My part: the rework that makes it work with any agent, and a multi-format exporter on Node.js, Playwright and pptxgenjs, with PPTX as images or as editable text.
The repositories
Questions people ask
- Yes. Four MIT-licensed projects on GitHub under nugrahalabib: AgentBuff Co-Founder, FinFreedom, Linky and AgentBuff Presentation Skills, all built to be used by AI agents as well as people.
- The Model Context Protocol is an open standard that lets AI agents call an application's functions as tools. AgentBuff Co-Founder and FinFreedom expose their features through MCP, with scoped tokens and, in FinFreedom, a human confirmation before any write.







