Mole: A Terminal-Based Deep Research Agent with Budget Enforcement and Verified Quotes

Introduction to Mole

Mole is a newly released deep-research agent designed for terminal use. It aims to provide a more controlled and verifiable approach to AI-driven research by integrating features like budget enforcement, source verification, and local data processing.

Key Features and Functionality

The agent decomposes user questions, searches for information, reads sources, extracts claims, and then cross-references these claims with their original text. It also identifies contradictions and generates an answer with citations. Mole operates by reserving and settling every model call against a predefined budget, ensuring that costs do not exceed the user-set limit.

A core feature is its quote verification system: any claim whose quote does not appear verbatim in its source is discarded. This ensures the accuracy and traceability of information presented in the final report. If a claim is later found unsupported, it is marked as such rather than silently removed.

Data Privacy and Local Processing

Mole prioritizes data privacy by processing local data, such as CSV files or folders, directly on the user's machine. Only aggregated results (counts, means, test results, buckets covering at least five records) are sent to external models, preventing sensitive raw data from leaving the local environment. Users can also view exactly what data leaves their machine.

Technical Implementation and Accessibility

The tool runs as a single static binary and utilizes the user's own API keys for model access. It supports Linux and macOS (amd64 and arm64 architectures) and can be installed via a curl script or Homebrew. Mole also supports the MCP protocol, allowing coding agents to drive it either by providing questions or by using Mole in toolkit mode for specific reasoning tasks.