Caveman

Founded by Julius Brussee

Cut token costs and compress context for AI coding agents and custom applications

Pricing model

Free Tier

Rating

★ People love it

Time to value

Quick Setup (< 1 hour)

You can use Caveman to shrink the token footprint of AI coding agents and custom applications without losing context. It compresses terminal logs, test results, JSON payloads, and diffs before language models process them, keeping the original text available on disk for recovery when needed. The middleware also trims verbose explanations from model responses so you receive direct answers. Additionally, it provides model routing to assign tasks to the least expensive model capable of completing the work, alongside cloud dashboards for monitoring team spending.

What Caveman does

Compress tool outputs and context windowsRoute queries to the most cost-effective language modelTrack API token expenses by agent and developerRecover full uncompressed tool outputs on demandTest prompt modifications against historical agent interactionsIntegrate context compression middleware into Python applicationsMonitor failed requests and debugging traces across agent workflowsCompresses terminal logs, JSON data, and git diffs before model processingPreserves original uncompressed context on disk for complete recoveryTrims conversational filler from model responses to lower output tokensRoutes individual tasks dynamically to cheaper models based on difficultyIntegrates via middleware into TypeScript and Python agent codebasesTracks agent token expenditure and costs across team members in a central dashboardSimulates and tests prompt modifications on past requests before deployment