Privacy policy

Effective date: May 14, 2026
Last updated: May 14, 2026

This page explains how Ambient Labs, Inc. (“Ambient,” “we,” “us”) handles your information when you use our website, our app, and the integrations you choose to connect. It's long because we'd rather over-explain than under-explain. If you only want the headline: we read what you point us at, we don't train models on your stuff, and we delete what you ask us to delete.

Questions are always welcome at support@ambientapp.ai.

1. The short version

Ambient is a personal AI that reads, drafts, schedules and follows up across the tools you already use. To do any of that, it needs access to those tools and a memory of how you work. The rest of this page is the careful version of that sentence.

Who we are. Ambient Labs, Inc., a Delaware corporation, headquartered in the United States.
Who this applies to. Anyone who visits our site, signs up for an account, or connects a third-party tool to Ambient. Adults only — we don't serve users under 16.
What this doesn't cover. The third parties you connect Ambient to. Their privacy policies govern what they do with your data on their side.

2. What we collect

2.1 What you tell us directly

Account basics from your identity provider (today that's Google) — name, email, a unique ID, plus a profile photo if you upload one. We never store passwords for Ambient itself — we authenticate you through your identity provider, every time. If you pay us, Stripe handles your card; we see the last four digits, brand, and zip — never the full number. Anything you send to support (emails, screenshots, recordings) we keep so we can keep helping.

2.2 What the connected tools share, with your blessing

When you authorize a tool via OAuth, Ambient can read what you let it read. Today that means parts of Google Workspace, plus the long tail of integrations you choose to connect, on the scopes you approve:

Gmail — message bodies, subjects, senders, threads, labels, drafts, attachments and signatures.
Calendar — events, attendees, locations, conferencing links, attachments and availability.
Contacts — the contact records you authorize.
Drive, Docs & Sheets — the documents you point Ambient at, including their text, metadata, comments and revision history.
Tasks — lists, items, due dates and statuses.

You can revoke any of this at any time, inside Ambient or directly with the provider. Revocation stops new collection and may break the features that depend on it.

2.3 The memory Ambient builds about you

We keep three quiet layers of memory so the assistant feels personal. They're stored as encrypted records in a PostgreSQL database with the pgvector extension — the same database that holds your account. Embeddings power semantic recall; the raw text stays plainly inspectable in your settings.

Semantic — facts: your contacts, your tone, your projects, your preferences.
Episodic — patterns: when you reply, what you open first, the rhythms of your week.
Procedural — workflows: the end-to-end sequences Ambient has learned to repeat for you.

You can browse, edit and delete any memory entry from settings. Deletion in Ambient doesn't delete the underlying source data in your inbox or docs (that's not ours to remove) and may reduce the quality of future drafts.

2.4 Usage, device and log signals

When you use Ambient we automatically capture device type, browser, operating system, language, time zone, IP address, an approximate (city-level) location, plus the usual telemetry: which features you use, latency, error events. We don't collect GPS-level location.

2.5 Cookies, locally and lightly

We use cookies and local storage to keep you signed in, remember preferences and protect against abuse. Inside the app we do not run third-party advertising trackers. On marketing pages we use a small number of analytics tools, and on EU/UK pages we ask first.

2.6 Things other people tell us about you

Sometimes we hear about you from elsewhere: identity providers when you sign in, your workspace admin if your company adopts Ambient, and a few analytics partners that tell us which marketing channels actually work.

2.7 How we use product analytics

We use PostHog (server-side ingest only — no PostHog code runs in your browser) to understand how the product itself is being used. The events we collect are limited to categories like:

Engagement signals — when you sign up, start a session, create a thread, send a message, hit a usage streak, or return after being inactive.
Skill / tool usage — which skills the agent invoked, whether they succeeded, how long they took, the model used. We tag each skill with a coarse use-case (email / calendar / notes / writing / research / etc.) so we can see what people mostly come here for.
Friction signals — clarification rounds, message rephrases (computed as a similarity score, never the text), tool failures, agent apologies, and abandoned conversations.
Onboarding flow — which step you're on, how long each step takes, where people drop off, which nemesis they pick in onboarding v2.
Retention & aha-moments — D1 / D7 / D30 anniversaries, time-to-first-value, churn / recovery flags.
Monetization lifecycle — paywall views, conversions, trial starts / expirations, subscription churn.
Cost & latency — model name, token counts per turn, p50 / p95 / p99 response times.

What we do not send to PostHog: the contents of your messages, email bodies or subjects, recipient addresses, attachment filenames, search queries, or any URL you visit through the agent. The analytics pipeline runs every event through a denylist that strips those fields before anything leaves our servers. We also suppress the entire event stream for threads you've flagged incognito and for individual messages you've marked “exclude from memory.”

Because no message content is collected, this analytics tier doesn't have a separate opt-out — the only data points are counts, buckets, timings, and the small set of categorical tags listed above. Your full Erasure / Export rights under §11 still apply: if you ask us to delete your account, we delete your PostHog events alongside everything else.

3. How we use any of it

To run the assistant. Authenticate you, draft, summarize, schedule, search, take action when you ask.
To make it feel like yours. Build the three memory layers above so Ambient's outputs sound like you.
To keep it safe. Detect abuse, prevent fraud, investigate incidents, enforce our terms.
To make it better. Diagnose bugs, measure performance, ship new features. This is distinct from model training — see §5.
To talk with you. Transactional emails, security and billing notices, support replies, occasional product news you can unsubscribe from.
To stay legal. Tax, accounting, audit, lawful requests, dispute resolution.

4. Legal grounds (for EU / UK / Swiss readers)

If you're in the EEA, the UK or Switzerland we rely on these bases under the GDPR / UK GDPR / Swiss FADP:

Contract — to deliver the service you signed up for.
Consent — for marketing to non-customers and non-essential cookies. You can withdraw consent any time.
Legitimate interests — for security, product analytics, light marketing to existing users.
Legal obligation — tax, regulator response and the like.

We don't make solely-automated decisions with legal effect on you. We don't knowingly process sensitive categories. If your mail or docs happen to contain such categories, we process them because that's the content you asked us to handle.

5. AI models, training and sub-processors

We use a mix of models we run and large-model providers like Anthropic, OpenAI and Google. The important promise: we do not train foundation models on your content. Not your prompts, not your inbox content, not your memory, not your outputs. We contractually require our model providers to do the same. We may use aggregated, de-identified telemetry — the kind that can't be traced back to you — for evaluation and safety.

We rely on a small set of vendors to deliver Ambient: model providers, cloud infrastructure, payment processing (Stripe), email and support tooling, and product analytics (PostHog — see §2.7). Our integration infrastructure is SOC 2 Type II and ISO 27001 certified. The full sub-processor list lives on our compliance page; we notify enterprise customers before adding new ones, as required by our DPA.

One caveat about AI outputs: they can be wrong, incomplete, or made up. Every action lands in your approval ledger first — please review anything important before letting Ambient send it.

6. Who else sees your data

The people you ask us to email or invite. When you tell Ambient to send a message, schedule a meeting, or share a doc, we transmit what's needed for that action.
Our sub-processors. Under contracts that restrict their use of the data to providing services to us.
Your workspace admin, if you use Ambient through an organization.
Counterparties in a transaction (a merger, acquisition, financing or sale), under confidentiality. We'd tell you if control of your data was about to change.
Authorities when we're legally required, or to protect safety, rights or property. We push back on overbroad requests.

We don't “sell” personal information, and we don't “share” it for cross-context behavioral advertising.

7. Connected accounts & Google APIs

When you connect a tool to Ambient you authorize specific scopes — visible on the consent screen at the moment of connection. Ambient is OAuth-only for connected accounts; we never ask for, see or store passwords for the services you connect. Our use of Google user data follows the Google API Services User Data Policy, including its Limited Use requirements: we use the data only to run the user-facing features of Ambient, never for advertising, and humans don't read it except with your explicit consent, for security work, or where the law requires.

8. Sending data across borders

Ambient is based in the United States and we process information there and in countries where our service providers operate. When we move personal data out of the EEA, UK or Switzerland we rely on the European Commission's Standard Contractual Clauses, the UK International Data Transfer Addendum, the Swiss FDPIC-recognized SCCs and supplementary measures (encryption in transit and at rest, access controls, vendor due diligence).

9. How long we keep it

Account info — while your account is active, plus up to 30 days after deletion in primary systems and up to 90 days in encrypted backups.
Connected content — only what we need for the workflows you've enabled, and otherwise per your settings.
Memory — until you delete the entry or your account.
Logs and telemetry — typically up to 13 months.
Billing and tax — for as long as the law requires (commonly seven years).
Marketing — until you unsubscribe, plus a brief suppression window so we don't accidentally email you again.

10. Security

We protect your data with the boring, important things: TLS in transit, encryption at rest, least-privilege access, SSO and MFA for our staff, network segmentation, vendor reviews, audit logs, secure-development practices and an incident-response runbook we actually rehearse. Our integration infrastructure is SOC 2 Type II and ISO 27001 certified. Connected accounts are OAuth-only; we don't store third-party passwords. Memory data lives in an encrypted PostgreSQL database with the pgvector extension, scoped per-user.

No system is perfect. If something happens that affects you, we notify you and the regulators required by law. You can help: keep your Google account on MFA or a passkey, patch your devices, and be picky about what scopes you grant.

11. Your rights

Depending on where you live, you may have some or all of these rights. We honor them regardless of location, to the extent we can verify you're you.

Access — a copy of what we hold about you.
Correction — fix anything inaccurate or incomplete.
Deletion — ask us to remove your data, subject to legal exceptions.
Portability — a structured, machine-readable export.
Restriction — ask us to limit our processing.
Objection — to legitimate-interest processing and direct marketing (always honored).
Withdraw consent — at any time, without affecting prior lawful processing.
Complain to a regulator — we'd love a chance to address it first, but the door is yours.

To exercise any of these, email support@ambientapp.ai or use the in-product privacy controls. We respond within 30 days (longer for complex requests, with notice). California, EEA, UK, and other U.S. state rights are covered through this same process.

12. Kids

Ambient isn't for children under 16, and we don't knowingly collect their data. If you believe a child has given us information, write to us and we'll remove it.

13. Updates to this policy

We update this from time to time. If we make a meaningful change we post the updated version here and notify you in-product or by email at least 30 days before it takes effect, unless the law requires something sooner. The “Last updated” date at the top always tells you when the current version went live.

14. Get in touch

Ambient Labs, Inc.
Email: support@ambientapp.ai