Lajos Deme Releases Mole: A Terminal-Based Research Agent with Strict Budget Controls
The open-source tool leverages the Model Context Protocol to provide verified research while preventing API cost overruns and data leakage.
Developer Lajos Deme has released Mole, an open-source deep-research agent designed for the terminal. The tool aims to eliminate common failures in autonomous AI research, specifically targeting budget overruns, source hallucinations, and data privacy risks.
Written in Go and distributed as a single static binary, Mole introduces a rigorous enforcement system for API spending. Unlike many autonomous agents that can enter unpredictable loops, Mole implements a budget system where model calls are reserved before execution to prevent cost overruns. According to data shared by Deme on Hacker News, the tool has achieved a measured budget overshoot of 0%.
To combat the issue of "hallucinations"—where AI agents provide confident but incorrect information—Mole employs a dedicated verification process. The agent extracts specific claims and checks them directly against the source text to identify contradictions. This ensures that the final output is grounded in verified data rather than model-generated assumptions.
Solving the Agentic Research Gap
The release arrives as a response to a growing trend of "deep research" AI tools, such as Elicit or GPT Researcher. While powerful, these systems often struggle with unpredictable API costs and the tendency to jumble sources. Deme noted that using agents for research is often frustrating when they "blow way past budget" and sound confident without providing the best possible answer.
Beyond cost and accuracy, Mole addresses the critical issue of data sovereignty. The tool features a privacy boundary for local data, enabling the analysis of local files, such as CSVs, without the data leaving the user's machine. This allows professional users to perform deep analysis on proprietary datasets without exposing sensitive information to LLM providers.
Integration and Industry Impact
By exposing its functionality via the Model Context Protocol (MCP), Mole is designed to be driven by other coding agents, making it a modular component in a larger developer workflow. This integration allows developers to plug a reliable, budget-capped research engine into their existing AI toolchains.
For the broader industry, Mole represents a shift toward "constrained autonomy." By prioritizing strict budget and privacy controls over unrestricted exploration, it targets a professional demographic that requires the efficiency of AI research but cannot risk uncontrolled spending or the leakage of proprietary data.
What to Watch
As Mole is open-source, its adoption will likely depend on how the developer community integrates it with other MCP-compatible tools. Future updates may expand the types of local data it can process or refine its verification algorithms to handle more complex contradictions in source material.