Quotient Labs
Fermat by Quotient Labs is a context-compression layer for Claude Code that reduces agent token spend by 34–51% without sacrificing output quality. It works through a local wrapper and lightweight proxy, requiring zero configuration changes to existing workflows. In benchmark testing across 15 paired sessions, Fermat cut costs by 47% on average while improving quality scores from 0.83 to 0.92.
- Developer Tools
- Artificial Intelligence
- Software Only
Funding
Raised to date
$500KRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
Founders
Founder details are not available yet.
Product
Problem
Claude Code sessions consume large amounts of tokens as context compounds across extended multi-turn interactions, driving up agent costs significantly. Users pay for repeated reads, redundant tool outputs, and bloated conversation history—often 40-60% more than necessary—without any corresponding quality benefit.
Solution
Fermat provides a context-compression layer for Claude Code that runs through a local wrapper and lightweight proxy, requiring zero configuration changes to how users prompt, use tools, or review diffs. The system applies three core methods: semantic context pruning to eliminate redundant information, optimized agent tools that do more per call and reduce repeated reads, and quality-preserving compression applied before cache writes with fail-soft passthrough when compression is unsafe. Fermat supports both the Claude Code CLI and the official VS Code extension. In benchmark testing on the SessionBench suite using SWE-Atlas tasks, Fermat delivered a 47% average cost reduction across Go, TypeScript, and Python bundles while actually improving quality scores from 0.832 to 0.915.
Target Audience
Software development teams and individual developers who use Claude Code extensively and want to reduce agent token spend without sacrificing output quality, particularly those running long, multi-ticket coding sessions.
Features
- Semantic context pruning that eliminates redundant or low-value context from the conversation before it reaches the model, while preserving recent instructions, tool relationships, and task-critical state
- Purpose-built Search and Edit file-access tools that do more per call, reduce repeated reads, and keep edits tied to files the agent actually inspected
- Quality-preserving compression applied before cache writes, with fail-soft passthrough when compression is deemed unsafe
- Local wrapper and lightweight proxy architecture requiring zero configuration or changes to existing prompts and workflows
- SessionBench evaluation harness measuring cost and quality over realistic multi-ticket coding sessions rather than one-shot tasks, with K=5 paired runs to account for 40-60% per-run cost variance
- Supports both Claude Code CLI and the official VS Code extension, plus Anthropic and Vertex AI (Google Cloud Agent Platform) backends
- Org-level support with centralized access control and billing