Mole: Deep-Research Agent with Enforced Budget and Verified Quotes for Your Terminal
Mole is a deep-research agent that runs in your terminal. It decomposes a question, searches the web, reads sources, extracts claims, checks each claim against the text it came from, and writes an answer with citations. Unlike chat-based agents, it enforces a strict budget, verifies every claim, and keeps your local data private.
Enforced Budget
Every model call is reserved against a budget before it happens and settled after, using a ledger with non-negative constraints in the database schema. The --usd 0.50 flag means the run stops at fifty cents. Measured overshoot across the test corpus is 0%.
Verified Quotes
Every claim carries a quote that appears verbatim in the source page. Claims that don't pass this check are discarded at extraction. Surviving claims can be re-read against their source, and unsupported ones are marked in the report.
Local Data Privacy
Point mole at a CSV or folder, and it analyzes the data without the contents leaving your machine. The model chooses a hypothesis template and column names; mole runs the SQL locally. Only aggregates (counts, means, test results, buckets covering at least five records) are allowed back. mole crossings shows exactly what left your machine.
Installation
Install via script (Linux/macOS):
curl -fsSL https://raw.githubusercontent.com/lajosdeme/mole/main/install.sh | shOr via Homebrew:
brew install lajosdeme/mole/moleArch Linux (AUR): yay -S mole-research-bin or mole-research. Debian/Ubuntu: download the .deb from the releases page. From source (Go 1.25+): go install github.com/lajosdeme/mole/cmd/mole@latest.
Configuration
You need a search provider and a model provider. Keys live in ~/.config/mole/config.json with mode 0600 — never in environment variables. Mole supports most LLMs, including coding agents, subscriptions, and local models.
MCP Support
Mole speaks MCP, so a coding agent can drive it — either by handing mole a question and collecting the answer, or in toolkit mode, where the agent does the reasoning while mole supplies the non-model parts.
It's free and open source, runs as a single static binary with no runtime dependencies.
📖 Read the full source: HN AI Agents
👀 See Also
yburn: Tool to audit and replace unnecessary AI agent cron jobs
yburn is a Python tool that audits AI agent cron jobs and replaces those that don't need LLMs with standalone Python scripts. The creator found 58% of 98 cron jobs were purely mechanical tasks like system health checks and git backups.
Custom Output Styles Collection for Claude Code
A developer has created 13 custom output styles for Claude Code that modify the AI's behavior through system prompts. The styles include Roast for brutal code critique, Socratic for guided questioning, Breaker for adversarial testing, Ship It for pragmatic solutions, Paranoid for security focus, and TDD for test-driven development.
Torrix: Self-Hosted LLM Observability Without Postgres or Redis
Torrix is a self-hosted LLM observability tool that runs as a single Docker container backed by SQLite. Install with docker compose up; logs LLM calls via HTTP proxy or SDK — tokens, cost, latency, full traces, PII masking, cost forecasting.
NVIDIA SkillEvaluator Comes to ClawHub: Is a Skill Actually Better?
ClawHub now integrates NVIDIA SkillEvaluator to show whether a skill improves an agent before you install it. Patrick Erichsen's eval view compares baseline vs. measured lift across 300+ verified skills.