Show HN: Mole – Deep research agent for your terminal
Mole is an open-source, deep-research AI agent designed to tackle common frustrations with existing LLM-powered tools. It guarantees budget adherence, verifies every claim against its source, and ensures local data privacy. This addresses critical concerns for developers and researchers, offering a transparent and controlled approach to AI-assisted information gathering.
The Lowdown
Mole is a newly released, open-source deep-research agent that aims to provide a more reliable and transparent experience than existing AI research tools. It addresses key pain points such as unpredictable costs, unverified claims, and data privacy concerns by implementing strict controls and verification mechanisms.
- Enforced Budgeting: Mole guarantees that runs will not exceed a pre-set budget, with 0% measured overshoot across its test corpus. Every API call is reserved and settled against a ledger, ensuring financial control.
- Verified Claims and Citations: All claims generated by Mole are meticulously checked against their source documents. If a quote does not appear verbatim, it's discarded, and claims that survive are fully cited, allowing users to re-verify information.
- Local Data Privacy: The agent allows for analysis of local data (CSV, TSV, JSON, JSONL) without the data ever leaving the user's machine. It uses SQL for aggregations, ensuring only statistical summaries (not raw data) are exposed to models.
- Flexible LLM Integration: Mole works with various LLMs, including local models and OpenAI-compatible endpoints, allowing users to leverage their existing API keys or self-hosted solutions.
- Toolkit Mode: Beyond autonomous research, Mole offers a 'toolkit mode' where a coding agent can drive the research process, with Mole supplying the deterministic components like quote checking and data aggregation.
- Comprehensive Inspection Tools: Users can inspect sessions, trace call costs, timings, and analyze fetch failures, providing full transparency into the agent's operations.
- Honest Numbers: The project provides self-graded metrics on budget overshoot (0%), claim integrity (100%), citation accuracy (100%), and grounding rate (80%), emphasizing its commitment to reliability.
By focusing on transparency, verifiable outputs, and local data processing, Mole positions itself as a robust solution for developers and researchers seeking greater control and trustworthiness in their AI-driven research workflows.