I'm Nipurn Agarwal, founder of 100XPrompt — an AI coding-agent stack: a terminal-native CLI, an SDK for building your own agents, and a plugin platform, together doing 9,000+ npm installs a month. Around the core product I build the infrastructure that agentic coding needs — chitta, a local memory layer that cuts context tokens up to 143×; MCP servers used with Claude Code, Cursor & Codex; and research into LLM inference efficiency (turboquant — 5–6× KV-cache compression). I also work across voice AI and LLM infrastructure, mostly in TypeScript, Python and Go.
100XPrompt — an AI coding agent that plans, edits, reviews and ships code from your terminal. Multi-model, with hooks, skills, sub-agents and MCP support.
- 🧠 chitta — permission-aware, permanent local memory for AI coding agents. Remembers across sessions, cuts context tokens up to 143×, and spends zero LLM tokens to store or recall.
- 🔬 turboquant — 5–6× KV-cache compression for LLMs with minimal accuracy loss.
- 🔌 duckduckgo-mcp — free, unlimited web-search MCP server. No API keys. Search, crawl & research with AI-ranked results.
- 📄 ilovepdf-mcp — MCP server for the iLovePDF / iLoveIMG API. 20+ tools: compress, merge, split, convert, watermark, OCR.
- 🗣️ Voice AI — real-time voice assistants and Indic speech model fine-tuning (Moshi / LiveKit).
| 🚀 100XPrompt | Full AI coding-agent stack — CLI + SDK + plugin platform, shipping at v0.1.27 |
| 📦 npm distribution | 9,000+ downloads / month across the CLI, SDK & plugin — entirely organic |
| 🧠 chitta | Local memory layer for AI agents — up to 143× fewer context tokens, zero LLM tokens to store/recall |
| 🔬 turboquant | 5–6× KV-cache compression for LLMs with minimal accuracy loss |
| 🔌 MCP tooling | Authored MCP servers used with Claude Code, Cursor & Codex |
| 🌐 Stack | TypeScript · Python · Go · Bun · GCP Cloud Run |
AI Coding Agents · Agent Memory & Knowledge Graphs · LLM Inference & KV-Cache Compression MCP Tooling · Voice AI · LLM Infrastructure & Security
