Private AI CRM

AI in your CRM.
Not your CRM in someone’s AI.

Lilly answers questions about your records, books meetings, sets reminders and finds the deal you half-remember. On hosted plans she runs on Qivity’s servers. Self-host and she runs on yours. Optional AI add-on on paid plans — not on the Free trial.

The architectural difference

Two ways a CRM can do AI. Only one keeps the data inside.

Every other CRM assistant

  1. Rep asks a question in the CRM
  2. Your records are packaged into a prompt
  3. The prompt leaves your infrastructure
  4. A third-party model provider processes it
  5. The answer comes back

Disclosable subprocessor. Metered per use. Residency becomes two separate decisions instead of one.

Qivity — Lilly

  1. Rep asks a question in the CRM
  2. A local model classifies what they meant
  3. The query runs against your own database
  4. Under that person’s own permissions
  5. The answer comes back — nothing left

Hosted: Qivity’s servers. Self-hosted: yours, including air-gapped. Lilly is an optional AI add-on on paid plans — ask sales.

What she is actually good at

A CRM assistant does not need to write your strategy memo. It needs to understand what a busy person meant and fetch the right thing. That is classification and retrieval — precisely where small local models are strong, and precisely where a hallucination would do the most damage.

  • Your own figures. Lead counts, pipeline value, breakdowns by stage, owner, region or source.
  • Filters inside the question. “prospects in Riyadh”, “leads from the Trade Show list” — and she tells you which filter she applied.
  • Reminders and to-dos. “next Friday”, “in 2 hours”, “every Monday” — drafted on a card you confirm.
  • Meetings. Booked on your real calendar and logged against the customer record.
  • Semantic record search. “What do we know about the pharmaceutical accounts in the south?” finds records that mean that.
  • Download the answer. Any record answer exports the full matched set to Excel.

The guardrails

The three things that make an AI assistant trustworthy

It can’t see what you can’t

Every data answer runs as a query under the asking user’s own permissions. A manager gets their reporting line. Being an administrator does not widen it — administering an account and reading a colleague’s pipeline are different things.

It quotes, it doesn’t compose

Documentation answers are extracted from the product docs with the section they came from, not paraphrased. You can check it.

It says “I don’t know”

When nothing covers a question she says so. A confidently invented pipeline number is worse than no answer, and every product that optimises for fluency eventually produces one.

Five questions worth asking any CRM vendor

Take these into your next demo. The hesitation tells you as much as the answer.

  1. Which companies’ infrastructure processes an AI request? Names, not categories.
  2. Is that named as a subprocessor in the DPA? If not, why not.
  3. Where does inference physically happen, and can I pin it to a region?
  4. Is AI usage metered separately from the seat licence?
  5. Can the assistant answer about records the asking user cannot open? Ask them to demo it as a junior rep.

The long version of this argument

AI CRM questions

Does Qivity’s AI send my data to OpenAI or any other provider?

Not to OpenAI, Anthropic or Google. On hosted plans Lilly runs on Qivity’s servers. Self-host and she runs on your organisation’s infrastructure.

What can the AI assistant actually do?

Answer questions about how the product works by quoting the documentation; answer questions about your own records — lead counts, pipeline value, breakdowns by stage or region; read your calendar; create reminders and to-dos, including repeating ones; schedule meetings; and find records by meaning rather than exact wording. Answers about records can be downloaded as Excel.

Can the assistant see records I’m not allowed to see?

No. Every data answer is executed under the asking user’s own permissions. A manager’s totals include their reports; everyone else sees their own records. Being an administrator does not widen this.

What happens when it doesn’t know?

It says so. Documentation answers are quoted, never invented, and when nothing covers a question the assistant says it does not know rather than guessing.

Does the AI cost extra per use?

Lilly is an optional AI add-on on paid plans — not a per-request meter, and not on the Free trial. Ask sales for the current rate. On hosted plans she runs on Qivity’s servers. Self-host and she runs on yours.

Watch it run with no internet connection

We will demo Lilly against a sandbox with egress blocked, and you can watch her answer anyway.