AI is genuinely useful for CRM work: answering questions about your pipeline, filling in missing fields, drafting outreach. The catch is that most AI tools do this by sending your customer data to a cloud model. For a lot of teams, handing contact and account data to a third-party AI service is exactly the thing they are not allowed, or not willing, to do.
A private, on-device AI takes the other path: the model runs on your own machine, and your data never leaves it.
What “on-device” actually means
The AI model runs locally, on your computer, not on a provider’s servers. There is no cloud inference step, so your Salesforce records are not sent to OpenAI, to Upcells, or to any third-party AI service. You download the model once, and after that it works offline.
That is the whole privacy story in one sentence: the data stays where it already is.
What it does (and what it does not)
Being honest matters here. A local model is a capable, practical copilot, not a frontier cloud model, and not magic. Used well, it is very good at concrete CRM jobs:
- Answer questions about your data. Ask things like “which opportunities slipped from last quarter?” and get an answer, using on-device retrieval over a local snapshot of your records.
- Enrich records. Point it at a field like a company Website and it reads the site and proposes values (for example, Industry) for you to review and apply. Pages it cannot read are flagged, not guessed.
- Draft outreach. Turn a few notes into a first-draft follow-up.
What it is not: a replacement for human judgement, or a system you should let write to Salesforce unattended. Every suggestion is yours to review before it is applied.
Why running it locally is the point
- Privacy. Customer data never leaves your machine, so there is no third-party AI processor to vet, and nothing to leak in transit.
- An easier security yes. “It runs on the user’s device and sends nothing to the cloud” is a much shorter conversation with IT and compliance than a cloud-AI data-processing review.
- It works offline. Once the model is downloaded, no connection is needed for the AI itself.
There is more on the trade-offs in on-device vs cloud AI for sales teams, the risk question in is it safe to put your CRM data into AI?, and the UK compliance angle in GDPR, AI, and customer data.
How Upcells does it
Upcells runs a local model on your machine as part of its Accelerator tier. The Assistant answers questions over local data pools, and data enrichment reads company sites and proposes field values for review. Because the whole app is local-first, your records flow directly between your computer and your Salesforce org, with the AI step happening entirely on-device. See the features for the full picture.
Related guides
- Is it safe to put your CRM data into AI?
- On-device vs cloud AI for sales teams
- GDPR, AI, and customer data: a guide for UK sales teams
Frequently asked questions
Does a private AI for Salesforce send my data to the cloud? No. An on-device AI runs the model locally, so your CRM data is not sent to any cloud AI service. You download the model once and it then works offline.
Is a local AI model as capable as ChatGPT? No, and it is worth being clear about that. A local model is a practical copilot for concrete CRM tasks (questions, enrichment, drafting), not a frontier cloud model. The trade-off is privacy and simplicity for raw capability.
Can I use AI on Salesforce data without a cloud service? Yes. A tool like Upcells runs the model on your own machine, so you get AI help with your CRM data without sending it to a third party.