Show HN: Mole – Deep Research Agent for the Terminal

Mole is a terminal-based deep research agent that decomposes questions, performs searches, reads sources, extracts and verifies claims, and then composes answers.

Key features include:

• Strict budget control: Every model call reserves its budget before execution, ensuring that the configured USD or token limit is never exceeded. The measured error margin is close to 0%.
• Citation-based verification: All claims are verified against the source text. Content not found in the original is discarded during the extraction phase or explicitly marked as unsupported in the final report.
• Local data privacy: When analyzing local CSV files or folders, the design ensures data does not leak externally by having the model generate SQL and return only aggregated results (statistics, averages, etc.).
• MCP support: Supports the Model Context Protocol, allowing coding agents to use Mole as a tool for research tasks.

Users can configure their own API keys (Anthropic, OpenAI-compatible, etc.) and search providers (Tavily, Brave, etc.) to easily run it as a single static binary.