PlatAtlasHomeHow starting works
From sign-up to your team deciding in its inbox, in an afternoon.
You pick one kind of request your team answers over and over, like "pay this invoice or hold it?". PlatAtlas learns how your team has answered it before, suggests the answer for each new one, and a person decides. Here is every step, and who does it.
Get started with your GitHub organizationWant to see it first? Try the sample inbox with a made-up company, right in your browser. The suggestions and the weekly check there are real; nothing you do is saved.
1. Sign up with GitHub
Each PlatAtlas organization is one GitHub account: your organization's, or your own if you start alone. Install the PlatAtlas GitHub App on it, sign in, and set it up; after that, its members sign in with GitHub. GitHub says who you are, and it is where your playbooks are kept: each one, with every email you forward to it, is a private repository in that account. PlatAtlas reads and writes them there and keeps no copy.
2. Point at the answers your team already gave
Past requests with the column that says what was decided: paid or held, approved or sent back, whatever your people wrote. Fifty is enough to start; more is better. Your steward connects them, read-only, from wherever they live:
Starting with your email? It needs no history. On your organization's page, Set up my inbox makes the playbook and gives you an address to forward to; your first answers become the history.
| Where they live | How PlatAtlas reads them |
|---|---|
| Files | An export: CSV, Excel, JSON, a folder of them, saved emails |
| Databases | PostgreSQL, MySQL and MariaDB, SQL Server, SQLite; MongoDB, DynamoDB, Elasticsearch, CouchDB: one read-only query |
| Your systems' APIs | A JSON API it reads page by page, or your system sends each new request in with its own key |
| The web | A table on a web page, or a Google Sheet |
| A mailbox | Forward it to your organization's PlatAtlas address: one rule in iCloud, Gmail or Outlook, set up from your organization's page with the steps for each. No password, and PlatAtlas keeps no copy: each email is saved only in your own GitHub organization. |
It only ever reads. Passwords stay on your side, never in PlatAtlas's files (a forwarded mailbox needs none); web pages are read politely, honouring each site's robots.txt.
Check your export now
Or go further: see the rules your team already follows and download your playbook, from the same file, in this browser.
Drop a CSV, Excel or JSON export here to see what PlatAtlas would make of it. It is read in this page and never uploaded.
Drop a file here, or .
.csv · .xlsx · .tsv · .json
3. It becomes your team's playbook
- It says what it found: which column holds the decision, and every column that looks like personal data (emails, phone numbers, addresses, account numbers). Those are left out unless you ask to keep them.
- Your past answers become the test. Before an AI model sees a single new request, it is graded on how many of your own past answers it would have got wrong, and what it costs. You keep the cheapest one that clears your limit.
- Nothing is decided alone. Every request reaches a person until your steward says otherwise.
- It's yours from the first minute: a folder of plain files, kept in your organization's own repository.
Today you can do this in your browser at /demo/start/, where Create makes the playbook's private repository in your GitHub organization, or with one command on your own computer.
4. Your steward's hour
Your steward is whoever knows how these requests should be handled. On one page, they write what each answer means in your team's words, next to the past requests the AI got wrong (this is where accuracy comes from); who may decide which answers; and what, if anything, may finish on its own. That last one comes with evidence from your own history, and nothing finishes alone until the steward presses Allow.
On our made-up demo company (60 past invoices, Claude Sonnet 5), the evidence said: “at 0.65 or above, 27 of 29 past paid cases would have closed with no person, and 0 of those calls were wrong.” Bank-detail changes and possible duplicates stayed with a person: only 5 past cases of each, too few to lean on.
5. Your team decides where it already works
| Where | What it's like |
|---|---|
| Web browser | Your organization's inbox on platatlas.com: sign in with GitHub, nothing to install. Works today. A one-click invite link by email works on your own server. |
| A plain weekly summary for the steward. Works on your own server; on platatlas.com it comes next. | |
| Outlook | Answer buttons inside the email itself. Built and tested; not yet running at a customer. |
| Microsoft Teams | A card in the chat or channel people already watch. Built and tested; not yet running at a customer. |
| Slack | A message with answer buttons. Built and tested; not yet running at a customer. |
| Your own portal | A small web interface to hand in requests, show a person's list, and record decisions. |
New requests arrive on their own. On platatlas.com, that is the email you forward. On your own server, it is also anything that can send a web request (Power Automate, Zapier, a script), or PlatAtlas re-reading the same database, folder or mailbox for anything new.
6. Every week, the check
The AI is graded again on your team's decisions, including every one made since. If it misses more than your limit, the check fails, says which answers it gets wrong, and nothing finishes on its own. This runs today on your own server; on platatlas.com it comes next. There, an AI agent on a playbook is checked against your team's answers once enough of them exist, or when someone who manages the playbook asks.
Where your data goes
| Question | Answer |
|---|---|
| Who holds our playbook? | You do: a folder of plain files in your own repository. We hold none of it unless you ask us to run it, on a machine we name to you. |
| What does the AI see? | Each request's details, minus the columns you left out as personal data, plus the question and your team's descriptions of the answers. Nothing else. |
| Which AI provider? | On your own server, the one you choose. With the Anthropic API, by default Anthropic does not train on commercial customers' inputs or outputs. You can also run an open model on your own hardware. On platatlas.com today it is TypeSafe, and only in a playbook where someone who manages it has added an AI agent; the page names TypeSafe before anything is sent. More providers later. |
| Our database passwords? | Stay in your environment. PlatAtlas stores a name like env:ERP_URL, never the password, and only ever reads. |
| Sign-in links? | Stored only in scrambled (hashed) form. A new invite stops the old link. |
| Certifications? | None yet. PlatAtlas is early, and a data-processing agreement is drafted for your counsel to review. |
Said plainly. On platatlas.com, starting from an export, forwarded email and the browser inbox run today. An AI agent that guesses at each forwarded email runs there too, once someone who manages the playbook adds one: its guesses stay hidden and are only compared with your team's answers until that person chooses to show them as suggestions. The weekly check comes next there, and runs today on your own server. Files, SQLite, web pages and APIs are read in tests on real data; the other databases and mailboxes connect through their standard drivers but have not yet been pointed at a customer's system. Outlook, Teams, Slack and the Power Automate connection are built and tested against Microsoft's and Slack's rules; none has run at a customer yet. A decision records the name of the person who made it and where; it is not yet signed with a key that proves it.
Get started with your GitHub organizationRather talk first? Tell us the request.
Looking for the old PlatAtlas console? It was retired on September 22, 2026. Nothing your organization stored has been deleted; for an export, write to craig@craigmerry.com.