Can I turn the AI off completely?

Yes, with one switch under Settings → General, and it is a real switch rather than a preference. With it off the ask boxes are not drawn, the paragraph at the top of a record is not there, a chart is built by hand instead, and an automation's model step is skipped with a sentence in its log while the rest of the run carries on. Nothing about your records changes.

Do I need my own OpenAI or Anthropic key?

No. The model is Kicklace's, metered against your account with a day's allowance that comes with the plan, and there is nothing to configure. If you also want to drive Kicklace from your own Claude, you connect it under Settings → Claude; that is your subscription and it is optional, because Ask Kicklace inside the app runs on the same set of tools.

Can the AI send an email to my customers?

No. It can write one, and what it produces is a draft on the person's record or in the composer with the link, for a person to read and send. Sending is one of the three acts that cannot be taken back, so where a tool wants one it writes down what it wants in a sentence and it waits on the Approvals screen until somebody decides.

What can the model see about my customers?

One record at a time: its name, its stage, the lists it is on, its own field values, where it first came from, and the last thirty rows of its timeline as a kind, a title, a day and who did it. Where an automation was started by an email or a support message it also sees those words. It never sees another record, the words of any other timeline row, or anything from another workspace, and nothing is remembered between runs.

How do I know what the AI did?

Every tool call is one row under Settings → Claude — when, who, which tool, whether it was a change, whether it worked — and every change it made is on the record's own timeline attributed to you, via Claude. An automation writes a line in its own log for each step, including the ones the model could not answer.

Does the AI move people between stages?

Not on its own judgement. The lifecycle is derived from events, and an automation you wrote and armed can move a record between stages of a pipeline you made. What the model does inside a run is bounded: fill in a choice from a field's own options, write one note, fill in blank fields, or answer the yes-or-no question that picks the branch.

Kicklace, an AI-powered CRM for micro SaaS

AI CRM for micro SaaS

Every AI feature in Kicklace, named the way the app names it, and every limit it runs under: off until you switch it on, a day's allowance, output checked before it runs, and a person on the end of anything that leaves the building.

What the AI actually does

Kicklace is an AI-powered CRM and customer platform for micro SaaS companies, and this page is the list of what that actually means. Every line below is a feature that exists, named the way the app names it, followed by the limits it runs under.

The short version: the model sets things up, answers questions about your own data, writes the automations you read before you arm them, and drafts. It does not send, it does not delete, and it is off until you switch it on.

Eight things, each one built

Named the way the app names them, so you can go and find each one.

Set it up by asking: Connect Claude to your workspace under Settings → Claude — from claude.ai, Claude Desktop or Claude Code — and ask it to do the setting up: make the list, write the welcome email, add the stage, add the field, build the view, connect your website's address, draft the automation. It works through the same set of tools every time, each one written down in your name on that same screen.
Ask Kicklace, inside the app: The same tools, without a Claude subscription. ⌘J anywhere opens a small box where the pointer is; there is an Ask button on a record's header and on a card in the pipeline, so “what should I do about this?” needs no name in it; and the search palette offers it when what you typed reads as a question. Each tool call is drawn as a card as it happens, with the one line it wrote about itself.
A chart from a question: Type a question and Kicklace writes a chart specification, checks every stage, list and field it names against your workspace, runs it, and shows you the answer. You can adjust a chart by asking again, edit its specification by hand, or keep one Ask Kicklace just ran.
An automation in plain English: Describe what should happen and it comes back as four blocks you can read — when, wait, only if, then — over your own stages, lists and templates. A draft with a problem in it cannot be switched on, and nothing runs until you arm it.
Four steps that ask the model mid-run: Inside an automation the model can be asked to fill in a choice from a field's own options, to write one note on the timeline, to fill in blank fields and only blank ones, or to answer a yes-or-no question that decides which branch runs. Each is one call, each writes only what it is allowed to write, and an answer that is not one of the options is refused and written in the log.
A paragraph at the top of a record: Where AI is on, every person and organization carries one paragraph and one next step, written from that record's own timeline and nothing else — no other record, and not the words inside any email. It says who wrote it and when, and there is an Update button. It is not activity: writing it does not touch the record's last activity, because a paragraph about somebody is not something they did.
A morning brief: Off unless you switch it on. At eight in your workspace's own timezone it works out what changed — who joined, who went quiet, what is waiting for your approval, which drafts are ready to send, which of your charts moved — and delivers one notice. The facts come from the database. The model writes at most two sentences over the top of them, is shown only those facts, and is told never to state a number or a name it was not given; the facts go out whether or not it answers. A workspace with nothing to say gets no brief.
Four things it notices: Also off by default, and none of them calls the model: people who read your pricing page twice this week and have never heard from you; people who signed up a week ago and never got started; people who cancelled and nobody wrote back to; a reply nobody answered in two days. Each arrives as one notice and a task per person, on that person's record, due today.

A chart from a question

Pick a question and the chart answers it, drawn by the dashboard's own code.

Ask for a chart:

Won this month

“How much did we win this month?” · closed by close date, compared with last month

$8,400

+65%$5,100 the month before

3 deals

Sample answers. In the app you can adjust any chart by asking, and Kicklace builds it from your own records.

Sample data, not a customer's. In a workspace any question works, and the chart it becomes is one you can keep, move, resize or edit by hand.

The leash

Six limits the code keeps, not six promises.

On by default, off with one switch: The switch is under Settings → General. With it off, every screen that would have asked the model does not draw the box, the paragraph at the top of a record is not there at all, and an automation's model step is skipped with a sentence in its log rather than failing the run.
A day's allowance: Every call is metered against the account, and the allowance comes with the plan and resets daily. The overnight pass that rewrites the paragraphs at the top of your records stops short of the whole allowance and keeps a reserve, so a night job cannot spend the day's calls before you have opened a screen.
Structured output, checked before it runs: The model never writes SQL and never writes code. It answers with JSON in a shape Kicklace defined, and every stage, list, field, template, member and event name in that answer is checked against your workspace's own vocabulary before anything is saved. A reference that does not resolve is a problem shown on the draft, not a guess.
Drafts for anything that reaches outside: The model can write an email; it cannot send one. What it produces is a draft on a person's record or a broadcast waiting in the composer, with the link, and a person presses send. The same is true of an automation: it is written as a draft and somebody arms it.
Approvals for what cannot be undone: Sending an email, deleting something, and revoking a key or disconnecting a door are the three acts that cannot be taken back, so the tools do not do them. They write down what they want in one sentence and it lands on Approvals in the sidebar, where a person decides. Everything else Claude does directly.
It takes no orders from your data: A note, a reply, the words somebody typed into a form: anything a tool hands back is data, and the model is told so. It never follows an instruction found inside a record; it tells you the record says it.

What the model can see

One record, and the message that started the run.

What a model step reads is one record: its name, its stage and pipeline, the lists it is on, its own field values, where it first came from, and the last thirty rows of its timeline as a kind, a title, a day and who. Where the trigger carries one, it also reads the message that started the run — the email or the support message — fenced and introduced as somebody else's words.

What it never reads is another record, the words of any other timeline row, or anything at all from another workspace. Nothing is remembered between runs. You do not bring an API key: the model is Kicklace's, and switching AI off in a workspace stops all of it at once.

The longer versions: Ask Kicklace, the connector Claude uses, the automation language, and what runs overnight.

Questions

Can I turn the AI off completely?: Yes, with one switch under Settings → General, and it is a real switch rather than a preference. With it off the ask boxes are not drawn, the paragraph at the top of a record is not there, a chart is built by hand instead, and an automation's model step is skipped with a sentence in its log while the rest of the run carries on. Nothing about your records changes.
Do I need my own OpenAI or Anthropic key?: No. The model is Kicklace's, metered against your account with a day's allowance that comes with the plan, and there is nothing to configure. If you also want to drive Kicklace from your own Claude, you connect it under Settings → Claude; that is your subscription and it is optional, because Ask Kicklace inside the app runs on the same set of tools.
Can the AI send an email to my customers?: No. It can write one, and what it produces is a draft on the person's record or in the composer with the link, for a person to read and send. Sending is one of the three acts that cannot be taken back, so where a tool wants one it writes down what it wants in a sentence and it waits on the Approvals screen until somebody decides.
What can the model see about my customers?: One record at a time: its name, its stage, the lists it is on, its own field values, where it first came from, and the last thirty rows of its timeline as a kind, a title, a day and who did it. Where an automation was started by an email or a support message it also sees those words. It never sees another record, the words of any other timeline row, or anything from another workspace, and nothing is remembered between runs.
How do I know what the AI did?: Every tool call is one row under Settings → Claude — when, who, which tool, whether it was a change, whether it worked — and every change it made is on the record's own timeline attributed to you, via Claude. An automation writes a line in its own log for each step, including the ones the model could not answer.
Does the AI move people between stages?: Not on its own judgement. The lifecycle is derived from events, and an automation you wrote and armed can move a record between stages of a pipeline you made. What the model does inside a run is bounded: fill in a choice from a field's own options, write one note, fill in blank fields, or answer the yes-or-no question that picks the branch.

Join the waitlist

We are onboarding one company at a time. Leave your email and we will let you know when your workspace is ready.