OrgGeniusAI assistant for Salesforce · Managed package
Ask your Salesforce org a question. Get an answer it can prove.
OrgGenius lives in the Lightning utility bar, answers in plain English (pipeline, accounts, cases, whatever your team actually asks) and cites the records behind every answer. When the org doesn't have it, it says so.
Enterprise Edition and above · Your own model key · Nothing to provision
Counted from the packaged metadata, September 2026.
What makes it different
The pitch is what it refuses to do.
Most of the risk in putting an AI on your CRM is what it might see, send, or change. OrgGenius answers each of those with a boundary you can verify, not a policy it was asked to follow.
It only sees what you can see
Every query runs as the person asking, with Salesforce itself enforcing their object permissions, field-level security and sharing rules. No service account, no privileged mode. If you can't open a record, neither can your assistant.
It never writes without a yes
Reading is read-only. The package ships no update and no delete. The few things it can create (a task, a logged call, a case reply) act on one record, are drafted in plain language, and are saved only after you confirm.
Your AI, your key
Anthropic, OpenAI or Azure OpenAI, chosen in settings. The key lives in a Named Credential in your org, so your data goes to your vendor under your contract and nowhere else. There are no OrgGenius servers in the path.
You decide what leaves
Permissions decide which records the assistant sees. Masking decides which values leave for the model: named fields, a pattern library for card and ID numbers, and a blocklist for codenames. Your own people still see everything they're entitled to.
For the admin
It answers questions about the org, not just the data in it.
“Why can't Sam see this record?” is the question a Salesforce admin actually gets asked on a Tuesday afternoon. Ask OrgGenius and it walks every route (profile, permission sets, sharing, View All), reports the per-record result, and names what would fix it.
No other assistant in this category reads the permission model at all. The org is the hard half.
Nothing to provision
There is no sync schedule because there is no second copy.
OrgGenius builds no index and runs no embedding pipeline. It searches Salesforce's own index, at query time, under the asking user's own sharing model. Nothing to host, nothing to re-permission, nothing to go stale.
Install
A managed package into your own org. Add a key, assign permission sets, pick a model. Under an hour, sandbox first if you prefer.
Ask
Every question becomes a query Salesforce runs as you. Totals are computed in Apex, not by the model, so the arithmetic is never hallucinated.
Nothing else
No Data 360, no connector to provision, no third data processor in your DPA chain. Your security team reviews one thing: a package whose every query runs as the asking user.
What it does
Built for the questions your team already asks.
Pipeline and accounts
Deals closing this month, big deals gone quiet, pipeline by stage, accounts near a location, meeting prep for the next call.
Cases and service
Open cases by account, recurring subjects, cases like this one, resolution context, what went wrong at a customer lately.
Inside emails and files
What did we tell them about delivery? Find the spec sheet. Summarise the contract on this opportunity. It reads the actual PDF.
The org itself
Who can see what, permission set breadth, users not logging in, recent setup changes, lockouts, login history.
Getting things done
Log a call from a dictated voice note. Draft a follow-up task, a meeting, a Chatter post, a case reply. Every one confirmed before it saves.
On your phone
Same assistant in the Salesforce mobile app, on record pages, with dictation, for the answer you need between meetings.
Where this is going
Chat with your CRM is now a commodity. The plumbing under it is not.
Anthropic's Salesforce in Claude plugin shipped in beta on 15 September 2026. It settles the “ask my org a question” layer, and it sharpens what OrgGenius is for: Sales and Service Cloud orgs that cannot justify Agentforce, and the work a plugin in a chat client is in the wrong place to do.
Next: a readiness scan that finds the data and configuration defects behind wrong answers, archive to your own S3 bucket or SharePoint at storage cost, activity sync as real Task and Event records, and document connectors without a meter. All inside the org, all under the person's own permissions. The assistant becomes the interface to those modules.
Readiness Scan
Built, in verification. 20 checks for any org, 26 more where CPQ is installed. Scored, ranked, every finding explained
Archive to your storage
Built against test storage. 100 GB on S3 for about $28 a year, against the four figures Salesforce Archive charges
Activity sync
Specified. Email and calendar as Task and Event records the org owns
Document connectors
Listed. SharePoint first, no credit meter in the path
Where the product is, honestly
Complete, in daily use, and not yet listed.
OrgGenius is in use in pilot orgs, with 678 Apex tests behind it and an evaluation harness that measures whether the assistant refuses what it should, scored on questions it has never seen.
It is not yet listed on the AppExchange. We are preparing for Salesforce's security review now, and beta installs are direct until it passes. The settings Overview above exists so your security team can read the posture themselves rather than take our word for it.