examples

Qirrel Pre-LLM Sanitizer & Preprocessor Examples

This directory contains real-world, production-focused examples demonstrating how Qirrel functions as a zero-token local pre-processor for LLM pipelines and AI Agents.

---

Example Guide

File	Scenario	Key Benefit
1-voice-transcript-sanitizer.ts	Preprocessing real-time Speech-To-Text (STT) audio outputs	Removes spoken fillers ("um", "uh", "you know"), stutters, and repetitions in <1ms before sending to LLM.
2-zero-cost-entity-prefilter.ts	High-volume support ticket & email lead pre-filtering	Extracts emails, phone numbers, URLs locally at 0ms and $0 cost to bypass LLM API calls on simple queries.
3-token-saving-prellm-pipeline.ts	Prompt token reduction & HTML payload cleaning	Compresses web payloads and strips junk text to cut downstream LLM input token costs by 30-50%.
4-agent-mcp-offloaded-parsing.ts	AI Agent tool offloading via Qirrel Agent Bridge	Offloads entity extraction and transcript cleaning to local deterministic tools so agents don't hallucinate.

---

How to Run the Examples

Using Bun (Recommended)

bun run examples/1-voice-transcript-sanitizer.ts
bun run examples/2-zero-cost-entity-prefilter.ts
bun run examples/3-token-saving-prellm-pipeline.ts
bun run examples/4-agent-mcp-offloaded-parsing.ts

Using Node & ts-node

npx ts-node examples/1-voice-transcript-sanitizer.ts
npx ts-node examples/2-zero-cost-entity-prefilter.ts
npx ts-node examples/3-token-saving-prellm-pipeline.ts
npx ts-node examples/4-agent-mcp-offloaded-parsing.ts