Quotient Labs

Developers and teams using Claude Code for agentic coding pay for large amounts of context and tool-call tokens, much of which is redundant or unnecessary for the task at hand. The site shows a sample session where vanilla Claude Code cost $60.62 and took 21 agent turns to complete a ticket. Teams on API billing pay directly for this waste, while those on flat-rate subscriptions burn through usage limits faster than needed. There is no indication users have a way to reduce this spend without altering how they prompt, use tools, or review diffs.

Fermat runs as a local wrapper and lightweight proxy alongside Claude Code, sitting between the CLI and the Anthropic API to rewrite outbound and inbound context before it reaches the model. It applies three techniques described on the page: semantic context pruning that keeps recent instructions and task-critical state intact while trimming older context, purpose-built tool calls that reduce repeated reads, and quality-preserving compression of tool outputs and assistant prose before cache writes, with fail-soft passthrough when compression is unsafe. It installs with a single shell command and requires no changes to prompting, tools, or custom skills, works with Anthropic or Vertex AI, and supports both the Claude Code CLI and VS Code extension. Pricing is $20/mo for unlimited usage or a metered option charging 10% of measured savings.

A limited-time 2-week free trial is offered through October 31.