Changelog
What changed in Rulebook in each release, newest first. A feature is "live-checked" only when it ran on real accounts or with the real AI model. Everything else is in development.
These results come from Brightloop, our demo company with synthetic data, on real HubSpot, Salesforce, Slack and Google accounts. They are not results from customers.
Plan changes in plain words, and approval rules
- The owner can change a plan in plain words, for example "also send it to the sales team". Rulebook shows the changed plan, and only the owner confirms it. It never removes an approval. In a test, 49 of 50 edits were right.
- A person can describe a repeated task in one sentence, and Rulebook makes the steps and the schedule. In a test, 28 of 30 descriptions gave the right steps.
- A run can pause and ask the owner a question, for example "Attach Brightloop Pricing Sheet - October 2026.pdf?", and go on after the answer.
- Approval rules for each level of risk. After 20 clean runs, Rulebook suggests that a review can stop, and it still checks 1 run in 10.
- Each receipt records the AI model and the cost of each step.
- Safer checks: Rulebook checks the text of PDFs for secrets before an AI model call, and it reads dates such as "tomorrow" and "next Thursday".
In developmentThese features passed tests with the real AI model, but they are not in the app for customers yet. Voice edits are in development too.
Trust checks, templates and purchase orders
- Each sentence and each value that the AI writes cites its source. Fixed code checks numbers, prices, dates and names, and a second model checks the meaning. A fact with no source becomes "Unknown: needs a human".
- Rulebook removes secrets such as keys and passwords from each AI model request.
- Rulebook fills your Google Docs and Slides templates and checks the result against your brand: colors, fonts, logo and tone.
- Rulebook reads a purchase order, also a scanned one, and compares it with the quote in HubSpot or Salesforce, field by field. Deal desk gets the result in Slack. Since M4, Rulebook does not send a scanned page or an image to an AI model, because it cannot check them for secrets yet. A purchase order with a text layer still works.
- One Slack card for each work item, with the proof in plain words.
Work from schedules, email and Slack, in HubSpot and Salesforce
- Work now starts from regular tasks on a schedule, from inbound email, from Slack request channels and from direct messages.
- Rulebook works in HubSpot and in Salesforce, and in Google Drive and Docs. It checks each change in the tool.
The first full loop: from a meeting to a CRM update with proof
- Rulebook reads a meeting transcript and finds the action items, each with the exact quote.
- The owner approves the plan in Slack. Rulebook updates HubSpot, saves a Gmail draft and keeps a receipt with the proof.
The foundation
- The base of the product: encrypted storage for the access tokens of your tools, one place for all AI model calls with the cost of each call, and Brightloop, our demo company with synthetic data.
- The first version of this website.
Rulebook is in early development. We start pilots with a small number of teams. Join the waitlist