Two minutes on what the AI features actually do — and what they
deliberately don't. For the full picture, see the full manual.
AI helps you work faster and get more from your notes, tasks and
OKRs — by finding action items, tidying up content, and highlighting what matters.
Find action itemsAI can scan your note text and find
potential action items.
Tidy up contentAI can help improve formatting, make text
clearer, and fix spelling and grammar.
Work across featuresAI touches Tasks, the Team board, Notes,
Calendar and OKRs — not just one page — helping you keep information organized and
actionable.
Stay in controlYou decide what to use. Review and apply
suggestions with one click.
Save time
Get to the important actions faster.
Turn insights into progress
Use AI to help you capture, organize and move work
forward.
Full quick start
Almost always your own key — two jobs are the exception. Add your own
key in Settings → AI — Claude (Anthropic) or OpenAI, your choice — and every
AI-enabled feature you can press a button for reads it from there. Until you do, a feature
that needs it is simply not on the screen, not a button that fails when you press it.
ChatGPT works differently from Claude: its requests go through STRIDE's server, still on
your own key, and your OpenAI account needs API credit — a ChatGPT subscription doesn't include
it. Claude stays the default; ChatGPT is a choice you make for yourself.
Two things are different: combining repeat notifications into one email, and folding
duplicate Message Center updates, both run automatically, server-side, on the organization
key STRIDE designates for its own background jobs, never yours — see What's actually where below.
Or your Organization provides the key. If an Organization Owner has set one AI key for
everybody, Settings → AI says Provided by your organization and names the provider, and
every AI feature uses that key, whichever of your organizations you are working in (the one you joined first when the
current one has none). Your own key, if you had one, is put aside for 30 days and
returned if the organization turns its key off.
What is recorded, and who sees it. STRIDE keeps a usage record of each AI request that goes through
its server: the feature, provider, model, outcome, token counts, an estimated cost and the time, never what was sent or
what came back. It is deleted after 90 days. On an Organization's key, the Organization's owners can see it for everybody
(Organization settings → Organization → Open AI insights), and if the Organization connects Datadog, the same record, with
your name, goes to the Organization's Datadog account. Requests on your own key in the browser are not in it.
It suggests. The app decides. A model reads a note, a comment, an
email or an imported file and points at what looks like an action item, a name or a
date — as words. Matching that word to a real person, a real date or a real
permission is this application's job, done in code, every time. Say a board has three
people — Sam Rivera, Bob Smith and Dana Okafor — and a note names only
"Bob": one match, so it's assigned. Two of them sharing a first name instead means
nobody gets assigned, with a note saying why, rather than a guess. A model here never
writes to the database, never picks a user, and never
invents a due date. A few things run with no button pressed — flagging a
possibly-duplicate objective, suggesting a spreadsheet's column mapping on a file STRIDE
doesn't recognize, and the two automatic jobs above — and each one is called out below,
not left for you to discover.
Where to find it. A capability on one page isn't a capability on
every page — see the table below for exactly where each one lives.
Ask STRIDE answers how-to questions. Press the Ask STRIDE button beside
Help, or ⌘/ on a Mac or Ctrl / everywhere else. It answers from STRIDE’s own
manual and can walk you to the control: Take me there rings it, and Take the tour walks every step
of a how-to. You press every control yourself. With a Claude key the answer appears as it is written: it shows Reading the STRIDE reference…, then Writing the answer…, then the words, and the steps and sources arrive when the answer is complete; leaving mid-answer stops it. It uses your own AI key; with no key at all it answers on STRIDE’s shared assistant (10 questions an hour) when a God Mode holder has switched that on, and otherwise searches the
manual instead. What it sends and keeps is in Ask
STRIDE, and what it sends. In the AI capabilities list it is the Ask STRIDE how to do
something card, under Help & guidance.
What's actually where
Tasks by Email
Create a personal task without opening STRIDE. Open the
AI Assistant panel from the top bar and you will find the address to mail, with a
button to copy it. Send from your primary sign-in email or a verified alternate on
your account — a message from anywhere else is not matched to you and nothing is
created. The subject and the body are read by your own AI Assistant, so it has to be
set up first; the task lands in your Personal Tasks, due today, with a short title (about eight words) and the detail in its description. It is never a team
task and never lands on a board.
The panel shows how many of your messages succeeded and how many failed.
Those count unique messages only — if the same one is delivered twice, or retried, it
is still one result, because the application records each message once and recognizes a
repeat. The address and the counts appear only when your organization has switched the
feature on for a team you are in; if you cannot see them, that is why.
Summarize an emailed task before you open it
The key points of the email are
already in the task when you get to it. The same read that turns your mail into a task also
writes the description: what is being asked, the dates and figures that matter, and what you
have to do — as a short list, not a copy of the message. It happens the moment the mail
arrives, so there is nothing to wait for and no button to press, and it costs no extra call to
your AI Assistant: it is part of the one read that was already happening. The task's name is
never changed by this, and the email itself is kept exactly as it arrived behind Read the
email.
Tasks
Import your tasks (Settings → Your data) reads a pasted
Word, Excel, CSV or text file and proposes tasks for you to review before anything is
written — an owner nobody named still lands on you, not on a guess, and an assignee that
doesn't match anybody on your roster is left Unassigned rather than picked for you.
Tidy formatting works
on a task's description and comments; the paper-plane button beside Watch and
Email, labeled Follow-up, opens the follow-up dialog, which drafts an email
from a couple of sentences of yours — it sends your assistant your notes, the task's title and the first
2,000 characters of the task's latest email with who sent it (on a forward, the email is replied to its
original writer). You name who it is to, and the chips add your signature, the email thread or the checklist
to the reply once it is drafted. With no key, it builds the email from your own words instead. Create › Compose (also the Draft an email quick action, or E) writes an email the same way with no task behind it: your assistant gets your notes, the names and your style, never a task title or an email, and it also suggests a subject. Switch Draft as to Slack (or use the Draft a Slack message quick action) and it does Draft a Slack message in your style instead: one short chat message, with a few emoji at most; you press Copy for Slack and paste it yourself, and your assistant gets only your notes and your style, never an address. Print with an AI summary…, on a task row's … menu, writes a summary from
everything on the task before you print it — its description, tags, the objective and key
result it supports, every linked note, attachments and the full comment thread, naming
who it's assigned to and who wrote each comment; you see the summary before anything prints.
Week Agenda writes a saved weekly brief from your own tasks, also reachable from the
Calendar page.
Team board
The same import wizard, reading onto a board instead of your own
list, plus Draft follow-up… and Print with an AI summary… on a
board task, same as Tasks above. Post a comment that sounds actionable and STRIDE offers a
Possible action item card underneath it, with no button pressed to ask — you decide
whether to keep it. Export this board… (or a single topic) writes a summary
into the export automatically while it's being built, from the tasks' status, dates and a
slice of recent comments; no one's name is sent.
Notes
Open a note and look for the AI assistant card in its own rail:
Find action items reads the note and parks what it finds on the checklist — a
suggestion, not a task, until you make one of it. This card stays visible even with no
key configured; it says so, and the button opens Settings → AI for you instead of
failing. The note itself has its own Tidy formatting with AI button, same as a task's.
OKRs
An objective, key result or quarter can carry a written insight —
prose about the progress, pace and health this application already calculated. Every
number in it is one you can already see on the bar beside it; the model is never asked
to do arithmetic. Re-evaluate OKRs in view, in the Focus strip's tools, reviews every objective
on screen in one pass and tells you how many requests it will make first. Exporting the
current view offers a Write the digest button for an executive-ready narrative. Inside Import OKRs, "Suggest a mapping with AI" proposes how spreadsheet
columns map to objective fields; on a file whose headers don't match a known layout, STRIDE
also checks automatically, with no button pressed, sending the column headers and a few
sample values. Adding or editing an objective also checks automatically, when you leave the
Title field, for a likely duplicate — advisory only, and you keep the final decision.
Running automatically, on a key STRIDE designates
Two jobs are different from every
other capability on this page: they run server-side, unprompted, on the organization key STRIDE
designates for its own background jobs rather than yours, reading colleagues' names, task titles and
comment excerpts — the same things STRIDE's operator account can already see. Combining repeat notifications waits a
short grace period before mailing you about an @mention and folds everything about one
subject into a single email rather than several; a later update is folded into your next
email instead of arriving as a repeat. Folding duplicate Message Center updates runs
hourly over your unread items and recognizes when several rows say the same true thing about
one subject, showing it once instead of several times — nothing recorded is deleted, only
shown once. Either one falls back to a plain rule, never silence, if the model is
unavailable.
Good habits
Your key lives in Settings → AI and nowhere else — it is never written
into a CSV export or a backup, and switching providers never carries one key's secret to
the other's field.
An ambiguous match is left for you to resolve, not guessed. That applies to a
name, a date and a due date alike.
A big import that's regenerating shows live progress, and if a regeneration ever
fails partway, whatever you already had stays on screen rather than being cleared first.
Removing your key turns a control off immediately — no reload, no reopening
Settings.
Most AI features only ever see your own writing. A few see more: printing a task with
an AI summary sends the full comment thread (with who wrote each comment) and who the task
is assigned to; the two automatic notification jobs above read colleagues' names, task
titles and comment excerpts on the key STRIDE designates, not yours. The Week Agenda's Regenerate is
sent from STRIDE's server on your key, so it finishes even if you leave the page. With
ChatGPT selected, every AI request goes through STRIDE's server on your own key rather than straight
from your browser.
Tasks by Email works with either provider: the AI summary of an email you send to your
Tasks by Email address runs on Claude or ChatGPT, whichever you have chosen, on your own key.
Words you'll see
AI-enabled feature
Anything that calls a model. Needs your own key in
Settings → AI to appear at all.
Suggestion
Something AI found — a possible action item, a possible owner.
Not written anywhere until you accept it.
Your own API key
A credential from your chosen provider's own console
(console.anthropic.com for Claude, or OpenAI's). Billed to your account and stored with
your account. With Claude it is used from your own browser, except the Week Agenda's Regenerate,
which STRIDE's server runs with it. With ChatGPT selected, requests go through STRIDE's server using
your own key, which the server reads for one request and never stores.
Want more detail — every AI button, per-provider setup, what each one
sends and doesn't? Open the full manual →