# Ratel

> **TL;DR.** Ratel is an open-source context engine for AI agents: it indexes an agent's full catalog of tools and skills, then injects only the handful relevant to the current turn instead of loading the entire toolset into every prompt.

- **Category:** AI / Developer Tools / Infrastructure
- **Stage:** emergent
- **Age:** 9 days
- **Origin date:** 2026-07-16
- **First detected:** 2026-07-17
- **Canonical URL:** https://earlyterms.com/term/ratel
- **Sources:** 7 primary URLs

## Definition

Ratel is an open-source context engine for AI agents: it indexes an agent's full catalog of tools and skills, then injects only the handful relevant to the current turn instead of loading the entire toolset into every prompt.

Giacomo and Roberto, who previously built agents for SaaS companies, open-sourced Ratel and launched it via [Show HN on July 16, 2026](https://news.ycombinator.com/item?id=48936491), publishing [benchmarks](https://benchmark.ratel.sh) claiming up to 91% token cuts across four models, with one production user reporting an 81% cost drop after replacing a 300-plus-tool agent's context in month one.

## Analogy

Named for the honey badger, famous for shrugging off threats far bigger than itself — Ratel does the same to bloated tool lists.

## Why it's emerging now

Giacomo and Roberto launched Ratel on Hacker News on July 16, 2026 — an open-source library that indexes an AI agent's tools and skills so only the relevant ones load per turn. Benchmarks claim up to 91% token cuts with near-parity accuracy, and the founders say one production user cut costs 81% in month one.

## Related terms

- *parent:* context-engineering
- *related:* context-rot
- *related:* tool-layer
- *competitor:* lean-ctx
- *related:* mcp-tool
- *related:* mcps
- *related:* model-context-protocol
- *related:* managed-agents
- *related:* agent-harness
- *related:* token-maxxing
- *related:* vault-context

## Sources

1. [Show HN: Ratel launch thread](https://news.ycombinator.com/item?id=48936491)
2. [Ratel — GitHub repository](https://github.com/ratel-ai/ratel)
3. [Ratel — company site](https://www.ratel.sh/)
4. [Ratel — BFCL v3 benchmarks](https://benchmark.ratel.sh)
5. [Ratel — documentation](https://docs.ratel.sh/)
6. [@ratel-ai/sdk on npm](https://www.npmjs.com/package/@ratel-ai/sdk)
7. [Ratel — LinkedIn brand post](https://www.linkedin.com/posts/ratel-ai_context-engineering-ai-activity-7472204505934118912-Joix)

---
_Generated by EarlyTerms · https://earlyterms.com/term/ratel_