Questili
Get in touch

Side quests

The code we give away

Libraries, agent tooling, and experiments, straight out of our own work. Apache 2.0 or MIT. Copy, fork, ship.

Questili on GitHub

Five projects. Two are experiments.

Licences
MIT, Apache-2.0
Languages
JavaScript, TypeScript, Python

  1. Tooling

    Jevis

    A second pair of eyes for coding agents.

    Hooks for Claude Code, Codex, and other coding agents that check prompts, tool calls, and stops against a library of hard-won lessons. It adds notes, refuses risky commands, or sends a turn back, and fails open when its model is unavailable.

    Clonegit clone https://github.com/GLDRoger/jevis.git

    GLDRoger/jevisMIT · JavaScript

    Source

  2. Library

    razorpay-recurring-upi

    Recurring billing for Indian SaaS, kept inside your app.

    A billing kit for Razorpay recurring UPI Autopay, cards, and e-mandates. Subscriptions, invoices, dunning, and entitlements live in your own database, with a storage-neutral core, a Prisma adapter, and a Next.js example.

    Installnpm i @questili/razorpay-recurring-upi

    Questili/razorpay-recurring-upiApache-2.0 · TypeScript · Alpha

    npmSource

  3. Application

    eevee-mcp

    A shared workspace for people and browser agents.

    A durable workspace where people and browser agents create, run, and improve reusable applets over WebMCP, the emerging standard for exposing a page's actions to agents.

    Clonegit clone https://github.com/GLDRoger/eevee-mcp.git

    GLDRoger/eevee-mcpApache-2.0 · TypeScript · WebMCP Challenge entry

    Source

  4. Experiment Prototype

    Citizen of India

    Life events, bills, and paperwork, in English, Hindi, and Kannada.

    A mobile-first hackathon prototype for the paperwork that follows life events. It uses fictional profiles and simulated responses, and it is not a government service.

    Clonegit clone https://github.com/GLDRoger/citizen-of-india.git

    GLDRoger/citizen-of-indiaMIT · TypeScript · Hackathon prototype

    Try the demoSource

  5. Experiment Research

    Parameter Golf Local

    The best language model that fits in two megabytes and thirty minutes.

    A local-hardware take on the Parameter Golf challenge: train under a 2,000,000-byte artifact cap in 30 minutes on one RTX 4090 or Apple Silicon machine, scored by compression.

    Clonegit clone https://github.com/GLDRoger/parameter-golf-local.git

    GLDRoger/parameter-golf-localMIT · Python

    Source

Writing

Lessons from building Avalon Flow, written down while they still stung.

More essays on nchandra.me

This board is a snapshot. The real action is on GitHub.

github.com/Questili

Some projects live on Nishchay Chandra’s own account, github.com/GLDRoger.

Earlier products

Alyna and Does The Math Work are still up, resting while Avalon Flow gets our hours.