guava
Menu
Book a call

Proof // Free to read

Guava Research

How I judge AI before I recommend it.
Two benchmarks that score models on the money they protect for a business, the AI setup I run Guava on, and my own AI usage, live.

See all four Book a 20-minute call

At a glance // From the leaderboards

Guava Research at a glance
Models tested15 so far, across two private benchmarks of 100 hidden problems each
Main Street BenchClaude Opus 5.5, GPT-6 Astra and GPT-6.1 Sol are tied at the top, protecting $85,989 to $87,578 of $100,000 on average across their runs, with overlapping run ranges
MSP BenchClaude Opus 5.5 protects the most: $883,722 of $1,000,000 on average across its runs
Open sourceguava-stack, free under the MIT license
LiveMy Codex, Claude Code and OpenRouter usage, refreshed every minute
Updated

The work

Open any of them to see how it works.

  1. 01 Benchmark // Small business

    Main Street Bench

    Which AI protects a small business’s money best? Each model is handed everything for a messy neighborhood bakery, with 100 hidden problems and $100,000 at risk. Its score joins four public benchmarks in the Main Street Index, ranked against cost.

  2. 02 Benchmark // Managed IT

    MSP Bench

    The same test for a managed IT provider: its tickets, billing, backups, client accounts, scripts and contracts, with 100 hidden problems that put $1,000,000 at risk.

  3. 03 Open source // MIT license

    guava-stack

    The AI setup I run Guava on. One hub in your own Notion holds your context, rules, skills and memories, and every assistant you connect reads it. It’s a small add-on to Notion’s setup guide, with 16 working rules, a free workflow review, and a model router for Claude Code and Codex. Nothing is sent to Guava.

  4. Live

    04 Live // From Elastic

    Live AI Dashboard

    My own Codex, Claude Code and OpenRouter usage as it happens: model requests, tokens by model and response times. It’s the observability I set up for clients, pointed at myself. Prompts and responses are never captured.

Want this for your own team? An AI Assessment ranks which of your workflows are worth automating, and every scope I write includes dashboards like mine.

Book a 20-minute call