Meet your twin.
An AI twin that works while you don’t. It watches your tools, surfaces only what needs you, and drafts the reply in your voice. External writes ask by default; you control any narrower autonomy per goal.
Private beta, onboarded personally. Private by default. Admins can’t read your chats; nothing sends without your approval.
Ava · now
Good afternoon. Since last evening I finished 6 things across your goals. 3 need your call before anything goes out.
Hi Maya, great question. For a 3-seat pilot, the Team plan is the right place to start. I can send the current checkout options, or we can hop on a call this week if that would help.
Sends from your Gmail, as you. Nothing goes out until you approve.
The other five ran clean. Inbox triage archived 14 newsletters, your morning briefing went out at 8:00, and the pipeline sweep found nothing else unusual.
Works across the tools you already use
Watch your twin work.
Faithful product previews with illustrative data. External writes are approval-gated by default.
A pricing question lands. Your twin drafts the answer in your voice and holds it for one tap.
Reply to Maya about pilot pricing
Northwind renewal has gone quiet
Usage dip flagged in #acme
Hi Maya, great question. For a 3-seat pilot, the Team plan is the right place to start. I can send the current checkout options, or we can hop on a call this week if that would help.
Sends from your Gmail, as you. Nothing goes out until you approve.
Proactive, and always in your control.
You hand it an outcome to own and the guardrails around it. It works in the background, with external writes asking by default unless you explicitly grant narrower autonomy.
01Connect
Link the tools you already use. Each call stays scoped to the user or workspace authority and grants recorded for that goal.
02It watches
It reads what comes in across your sources and surfaces only what needs you, with the reasoning attached.
03You approve
It drafts the work. External writes are approval-gated by default, with autonomy you can explicitly narrow per goal.
04It learns
Edits and explicit corrections become feedback. Successful goal runs can also refine a plain-language procedure for that goal.
Ava, your twin
Runs as you, within your guardrails
Guardrails
Built to be trusted with real work.
Private by default. No admin can read your chats or memory. Enforced in the backend.
Governed writes. External writes ask by default. Explicit goal autonomy never overrides hard risk and permission gates.
Permissions per fact. Checked at read time. Lose access, lose the answer.
Encrypted and audited. Encrypted in transit and at rest. Every run audited.
It works across the tools you already use.
Connect once, scoped to what you authorize. Every connector runs under your grants, and external writes follow the goal's approval policy.
Gmail
Reads and drafts email; sends under the goal's policy.
Google Calendar
Reads your schedule; preps briefs before meetings.
Slack
Reads channels you are in; posts under the goal's policy.
HubSpot
Reads deals and contacts; flags what is slipping.
Salesforce
Reads CRM records and performs granted actions.
Google Drive
Reads files you authorize; grounds answers in permitted sources.
Not a fleet of bots. One twin each.
Each member gets one proactive twin in the workspace, shaped by their own preferences and permissions. No fleet of agents to wrangle.
Questions, answered.
Can it do things without me?
Not unless you let it. External writes are approval-gated by default. Per goal, you can choose ‘Ask before acting’ or ‘Ask only for risky actions’; hard permission, destructive-action, and admin gates still win.
Is my data private from my admins?
Yes. Your chats and personal memory are yours. Admins manage the workspace, its roles, connectors and billing, but they cannot read anyone's assistant chats or edit their personal memory. It is enforced in the backend, not a toggle.
Which tools does it connect to?
Gmail, Google Calendar, HubSpot, Salesforce, Slack, Google Drive and other catalogued apps, added through Connections. Available reads and actions vary by connector and stay scoped to what you authorize.
How does it get better over time?
Edits and explicit corrections are retained as feedback without guessing which skill caused them. Verified runs feed the governed skill-learning pipeline, and each goal can distill its own successful approach into a reusable procedure.
Is it secure enough for enterprise?
It includes workspace isolation, audit trails, role-based access, SSO and SCIM support, and encryption in transit and at rest. SOC 2 readiness work is in progress. Confirm any certification, residency, or regulated-workload requirement with us before deployment.
Give everyone a twin.
Proactive, private, and governed. It learns how each person works and clears the busywork, set up in minutes.