Frequently asked questions
The questions people actually ask — grouped by the part of Pulse they are about. If you are new, start with Getting started; if you are rolling Pulse out to a team, the security and administration sections are the ones to read end to end.
Capability
Apps · Meeting Agent · Code Agent · Devices & voice · API
The basics
What is Pulse, in one sentence?
A workspace assistant that lives in one persistent timeline per person: it reads what you ask, plans the work, runs the long parts in the background, and brings results — and any approval it needs — back to the same thread.
How is it different from a chatbot?
Three things. It keeps working after you close the tab, because tasks run on a durable queue. It acts on your systems through connected accounts, under policy you control. And it has one thread, not a new session each time, so context and history accumulate instead of resetting.
Why only one chat? I want separate conversations.
You get separate strands without separate chats: topics file each line of work under its own heading inside the one timeline. That keeps context in one place for the assistant while still letting you jump to "the Halden renewal" from the Topics panel. See the timeline.
Does the assistant read my whole history every time?
No. Recent conversation stays in the working context, older spans are summarized, and durable facts move into memory, which is retrieved by meaning rather than replayed wholesale.
Can more than one person use the same timeline?
No — a timeline belongs to one person. Sharing happens at the level of the things in it: files, todo lists, and organization-wide apps.
The timeline
What can appear in the timeline besides messages?
Task updates, approval requests, questions from the assistant, connect prompts for accounts it needs, generated files, and reactions. They all share one thread — there is no second inbox.
What is the panel next to the chat?
Your answer to "what needs me, and what came out of this": Overview, Topics, Todos, Pinned, Files, Tasks, plus any panel an installed app contributes.
Some panels are missing from the row.
They are under More. The row measures itself against the space available and collapses what does not fit; app panels start there by default. Drag to reorder, or set the arrangement in Settings → Appearance.
Can my organization decide the panel layout?
Yes — an admin can set a default arrangement for everyone or lock it entirely. When locked, the customizer says so.
How do I find something from weeks ago?
⌘K (Ctrl+K), then narrow with All · Chat · Topics · Files. Chat search is hybrid semantic + keyword, and each result expands to show the surrounding conversation rather than just the matching line.
What does pinning do?
Keeps a message one glance away as the thread grows — decisions, specs, an attached file. Pinned messages have their own panel.
Tasks & automation
What is the difference between a message and a task?
A message is the conversation. A task is work the assistant hands to a background worker because it is long-running — research, document generation, multi-step integration work. Tasks report progress and results back into your timeline.
Can I ask for something to happen on a schedule?
Yes, in plain language: "Every weekday at 9:00, send me a digest of yesterday's support tickets." Pulse registers it as a scheduled task and runs it without further prompting.
Where do I see what is running?
The Tasks panel, and Settings → Tasks for the full history — every run, and the error on any that failed.
What happens to a task if I close my browser?
It keeps running. The queue is durable; results land in the timeline whether or not you are watching.
Does a scheduled task ask me for approval?
It can, and this is the part worth understanding: an unattended run has no composer, so the Manual/Auto switch does not apply to it. Whether a tool pauses is decided by policy. See autonomy and approvals.
Todos & check-ins
Are Pulse todos separate from my chat?
No — the same items. You add them, the assistant adds them out of a conversation, and both of you see one list.
What is a check-in?
An on-demand or scheduled review of what needs your attention. The Check-in button in the Todos panel asks for one immediately; recurring check-ins and reminders are configured under Productivity settings.
Will it notify me at night?
Not inside your quiet hours. They are part of productivity settings and apply to reminders and check-ins across surfaces.
Can I assign a todo to a teammate?
Yes, and lists can be shared. Comments live on the item.
Autonomy & approvals
What do Manual and Auto actually change?
They decide, for that one message, whether governed tools pause for you. Manual asks before each; Auto lets them run. Reads are never gated by this — Manual does not ask permission for the assistant to look at a file.
Does Auto stay on?
No. It returns to Manual after each send unless you opt into sticky mode, and the first time you pick Auto you get a one-time explanation to confirm.
Auto is on but Pulse still asked me. Why?
Something above the composer applies — most often an organization rule set to Ask first, which deliberately outranks Auto. The decision ladder names exactly which layer won.
What is "yes to all", and how far does it go?
It covers one tool, for one job — the current background run or the current chat turn — and retires with that job. It can never become a standing permission, and it never reopens a tool that is switched Off.
Where can I approve from?
In chat, from an email deep-link, or the Approvals view. A decision in any channel resolves it everywhere.
Do I see what the action would actually do?
Yes — the real arguments: the actual recipient, the actual command, the actual record. Approvals show the call, not a paraphrase of it.
Who can approve?
The owner of the work. Delegation, four-eyes approval and expiry are not available today.
Security & governance
Who decides what the assistant is allowed to do?
Layers, each answering one question: your plan decides what exists, an admin decides who gets it and under what rule, your connected account decides what it can reach, and the composer decides whether this message runs unattended. Security & governance documents all of them.
What are the three modes?
Automatic (runs), Ask first (pauses for approval on every call), Off (removed from the assistant entirely — it cannot call it or mention it).
What is a policy?
A named set of rules an admin assigns to everyone, to groups, or to individual people. It is Active from its first save; turning it Inactive keeps it stored while applying it to nobody.
Two policies cover me. Which one applies?
The narrowest. A policy naming you beats one naming your group, which beats one covering everyone. If two reach you at the same level — two groups — the stricter answer wins per setting, so the result never depends on which was edited last.
If an admin writes a narrow policy, do I lose everything else?
No. A policy is a delta: it changes only what it states, and everything it does not mention keeps coming from the baseline. An exception is an exception, not a blank slate.
What is a "ceiling"?
Each rule carries a default and how far a member may relax it themselves. Lock the ceiling and the rule is a mandate; open it and the default becomes a starting point. One sentence covers the whole model: tightening flows freely, loosening needs permission from above.
Can I write one rule for a whole app or server?
Yes — that is what the <scope>::<tool> grammar is for. app:acme-crm::* covers everything that app exposes; app:acme-crm::delete_* covers its deletion tools, including ones shipped next month. The editor tells you how many tools a rule currently matches before you save.
Why can't I write a rule against a specific MCP server name?
Because every member gets their own copy of a connected server, so the same server is twenty different ids in a twenty-person org. Scopes (app:, catalog:, custom) are derived from fields members cannot write, so one rule reaches every copy.
Can a member fake their way into a looser bucket?
No. Which bucket a server lands in comes from columns the install and catalog flows write, not from user input. The only route in is installing the app — itself governed by app audience and the self-service gates.
What are the self-service gates?
Four organization switches, independent of policy, each defaulting to on: whether members may edit their own Security tab, add personal MCP servers, install or create skills, and add sub-agents. Closing the skills gate also closes app self-install.
Can Pulse reach our internal network?
Not by default. Every model-influenced request is checked against a blocklist and re-checked after DNS resolution, redirects are not followed, and private ranges are refused. On-prem deployments that need an internal service must open two gates — an environment flag and a per-installation allowed-host list.
Where are credentials stored?
Encrypted, in four separate stores with no read-back path: the OAuth client secret on the installation, each member's token on their grant, org-shared installation secrets in their own column, and personal variables on the user. APIs return the names of configured secrets, never values, and secrets are redacted from error messages.
Is there an audit trail?
Yes — security policy saves, gate changes, and app state/mode/audience changes are recorded with actor, target, and before/after values. Combined with task run history and the approvals record you can reconstruct what was asked, who decided, and what happened.
How quickly does a policy change take effect?
Within about 15 seconds.
Can I try a policy without applying it?
Yes — leave it Inactive and read the rules first. A new policy also starts as a copy of the baseline, so making it Active changes nothing until you edit it.
Apps & integrations
What is an app, versus an integration?
An app is the unit of capability — tools, skills, MCP servers, sub-agents, a chat panel, settings. An integration is the account an app acts with on your behalf. Slack and Google are apps; your two Google accounts are integrations.
Do I have to set up integrations before I start?
No. Ask for the work; the first time an account is missing, Pulse says so in the thread and offers to connect. Completing it resumes what was paused.
The assistant says it cannot use a tool. What now?
Read the reason — Pulse distinguishes them. Missing account, expired connection, or two accounts with no default are all fixable from the card. Hidden by policy, or not covered by your plan, are not — no amount of connecting changes those.
Can I connect two accounts for the same provider?
Yes, with one marked default. If you never pick, Pulse adopts the one it was already using rather than interrupting to ask.
Why does an app keep asking me to reconnect?
A token refresh is failing. Reconnecting issues fresh tokens; if it recurs immediately the provider is refusing the refresh — usually a revoked authorization, a rotated client secret, or a scope change on their side.
Who can install apps?
Members can install for themselves when the organization allows self-service. Admins install org-wide, and can mark an app required so members cannot remove it.
What does uninstalling remove?
The capability the app created — its skills, sub-agents, MCP servers, incoming webhooks and listeners. It never touches user content: meetings, files, change sets and chat history stay.
Can we build our own app?
Yes, from a declarative JSON manifest — no code to host on the Pulse side. See Building apps, and App surfaces for the live-panel protocol.
Can an app show its own screen inside Pulse?
Yes — one panel per app, fetched from the app with your credential injected server-side, with buttons and forms that call back. Sub-pages live inside that one panel.
Meeting Agent
What does it do?
Sends a notetaker bot to your call. It joins, records, transcribes, and writes structured notes — TL;DR, key points, decisions, action items, open questions — into your timeline and your files.
Which platforms?
Google Meet, Zoom and Microsoft Teams. Meet works with a signed-in Google account or when the host allows anonymous participants; Zoom requires an authenticated session because its anti-bot checks only admit signed-in participants; Teams joins via the web client.
How do I send it?
Two ways. Paste a link and ask — "Join this meeting and take notes: …" — or let it work from your calendar. With calendar sync on, Pulse scans connected Google calendars for meetings carrying a Meet, Zoom or Teams link.
Will it join meetings automatically?
Only if you tell it to. The auto-join policy is Off, Ask (Pulse requests approval before each) or Auto. Default lead time is 12 minutes before start, configurable 0–60. If an Ask request goes unanswered by start time, a fallback setting decides whether to skip — the default — or join anyway.
Will people know a bot is in the call?
Yes. It appears in the participant list under a display name, by default "<your name>'s Pulse", which you can change. It may land in the lobby first and needs the host to admit it; it gives up after 10 minutes if nobody does.
Can I watch what it is doing?
Yes — the meeting card shows a live view refreshed every few seconds, the participant roster, a streaming activity log, and a Leave & note button that pulls it out early and finalizes notes from what was captured.
Does it ever leave on its own?
It does, so you never have to remember: when the meeting ends, when it has been alone for about 90 seconds, when a calendar meeting runs past its scheduled end plus a 10-minute grace period, or at the maximum duration cap (default 3 hours, configurable).
What do I get afterwards?
A summary message in the timeline, plus a notes file and a full transcript file in Markdown saved to your notes folder. The raw recording is off by default — enable keep recording if you want it.
Does the transcript say who said what?
No. It is a best-effort text of the meeting audio without speaker labels; the participant roster tells you who was present.
What if it could not record anything?
Pulse posts a plain note explaining why — never admitted, call ended before any audio. There are no silent failures.
Which plan includes it?
Enterprise. On other plans it appears in Apps with a Requires Enterprise plan notice. Admins can disable it workspace-wide, and it is off per user until enabled.
Can I manage it from chat?
Yes — "Turn on the meeting agent and set auto-join to ask, 10 minutes before start" works, as does "List my recent meetings."
Code Agent
What is the sandbox?
An isolated container with a real git checkout, where the agent edits files, installs dependencies, and runs the actual build and tests. Your machine is not involved.
Does it push code or open pull requests by itself?
No. Work becomes a reviewable change set — a diff artifact in Pulse. You inspect it and open the PR from there. Posting a review to GitHub always goes through an approval, and pushing to main is forbidden.
What else can it do besides writing code?
Review a pull request (reading metadata, changed files, review threads, CI checks and job logs, plus the code on the branch), resolve merge conflicts semantically and push the resolved branch, and write architecture or audit documents grounded in a code-intelligence graph of the repo.
Can it look at several repositories at once?
Yes — up to six additional repos cloned read-only alongside the writable primary one, producing one combined cross-project report.
Do files it creates appear in my workspace?
Only when delivered. Sandbox files stay in the sandbox; code changes arrive as a change set, documents are written into Files, and generated batches can be bulk-exported with their directory structure intact.
Which plan includes it?
Pro and above.
Files & artifacts
What can Pulse produce?
Markdown, PDFs, Office documents, diagrams and web artifacts, written straight into the workspace file system rather than pasted into chat.
Can I get an earlier version back?
Yes — revisions are kept, comparable, and restorable.
How is sharing controlled?
Per resource, with a note if you want one, under organization defaults that govern how work may leave the workspace.
Memory
What does Pulse remember?
Durable facts — preferences, decisions, project details, relationships — stored as discrete items rather than transcripts, so recall is precise.
Can I see and correct what it knows?
Yes. Every fact is listed, searchable, editable, and can be marked obsolete.
Is memory the same as the timeline?
No. Clearing conversation noise does not erase what the assistant learned, and vice versa.
Devices & voice
What is the desktop app for?
Work that belongs on your machine — local files, local tools, device-specific context. The agent dispatches a task to the device and results report into the same timeline.
Is local access governed?
Yes, and separately: terminal, filesystem and browser capabilities have their own rules, with writes and commands defaulting to Ask first.
What is voice mode?
The same assistant, spoken. It has your context already, can run tools mid-call, and leaves a transcript.
API & webhooks
How do I call Pulse programmatically?
Personal access tokens against /rest/v1/*. Tokens are stored hashed, shown once, carry an optional expiry, and can be revoked.
What are token scopes?
Eight coarse permissions — profile, messages read/write, files read/write, MCP management, webhook management, memory read. A request missing the required scope gets a 403 naming it. Scopes are an and on top of role checks: a webhook scope does not bypass the admin requirement.
My old token has no scopes. Does it still work?
Yes — tokens minted before scopes existed are grandfathered with full access. New tokens always carry an explicit list, and tokens minted on an app's behalf are never grandfathered.
What is an incoming webhook for?
Letting an external system start Pulse work: a deploy finishes, a ticket moves, a form is submitted. You describe what to do with the delivered event; every delivery is stored so you can see what arrived.
Administration: plans, seats, roles
What do the plans gate?
Capability. The Meeting Agent requires Enterprise; the Code Agent requires Pro or above. On a plan without them the app is visible in the available state with a Requires plan notice.
What happens if a plan lapses?
Tools already in place refuse politely — the assistant explains the feature is not licensed rather than failing cryptically. No connect prompt is raised, because connecting would not help.
How are seats counted?
Three separate counters: a member limit, Pro licenses, and Enterprise licenses. Pending invitations reserve a seat against the member limit, and none of the three can be lowered below what is currently assigned.
What is a group for?
Targeting a subset of the organization without naming individuals — app audiences, security policy assignment, and directory-group mapping all take a group. Membership is evaluated live, so someone who joins later is covered automatically.
Privacy & data
Does Pulse act with more access than I have?
No. It uses your connection and your permissions on the far side of every integration.
Can an admin read my chat?
Administration covers policy, apps, seats and audit — not the contents of another member's timeline.
Can we run Pulse on our own infrastructure?
Yes. On-prem deployments add an environment-based license, seat caps enforced from it, and the private-egress gates described above.
Still stuck?
- The Troubleshooting section of Security & governance covers refusals, read-only pages and missing tools.
- Per-area troubleshooting lives at the foot of Apps and Integrations.
- The in-product API Reference under Settings documents every REST endpoint.