PrivacyPick
Morse logo

Morse

7.4
out of 10
Visit Morse →
IP AnonymizationStripped before model provider
Storage ModelLocal, device-only
RetentionChats stay on your device
JurisdictionUnited States
Free TierLimited free tier
Telemetryalways-on
Established2026
Web
GPTClaudeGeminiGrokLlamaDeepSeekKimiQwenGLMMiniMax
Pro (monthly)
$19.00/mo
Pro (annual)
$17.00/mo
Max (monthly)
$49.00/mo
Max (annual)
$44.00/mo

Published November 14, 2026Updated November 14, 2026By Juan Martinez

In This Article

Morse is a browser-based AI chat from Watermark Labs, Inc. of San Francisco. It reaches more than 80 models from OpenAI, Anthropic, Google, xAI, DeepSeek and others, with image generation, role-play characters and a pay-as-you-go API next to the chat. The product opened to everyone on 11 October 2026 and its site says Y Combinator backs it.

Chats are stored encrypted in the browser, and Morse’s servers hold no message they can read. Each request is sent in one of three modes that decide what the model company can see. In the top mode, the browser encrypts the message so that only a hardware-isolated server run by the model company can read it.

Part of the code is public and checkable, the web app is not, and no outside party has audited any of it.

Privacy Architecture

Every request goes out in one of three modes, chosen in the model menu. In Anonymous mode the models come from OpenAI, Anthropic, Google and Meta. Morse sends the message under its own account, so the model company reads the message but never learns who sent it, and may keep it for a limited time to check for abuse under its own policy. In Private mode the models run on hosts such as DeepInfra and Novita AI that Morse says have agreed to keep none of the messages. Where a host confirms zero retention on each reply, Morse checks it and refuses any reply that does not carry it; otherwise it relies on the host’s commitment, and those agreements are not published. Both modes hide the sender from the model company (IP anonymization) and neither hides the message from it.

The third mode, E2EE, is the one that hides the message too. Before sending, the browser checks the hardware attestation of two servers: Morse’s own enclave on AWS Nitro, and the enclave run by the model company, which uses Intel TDX or AMD SEV-SNP processors with NVIDIA GPUs. An attestation is a statement signed by the hardware that a server runs a named piece of software. The browser then encrypts the message twice. The outer layer holds routing and billing details for Morse’s enclave, and the inner layer holds the message for the model company’s enclave. Morse relays the result and cannot open the inner layer. The model’s reply comes back encrypted for the browser alone. The enclaves belong to Phala, NEAR AI and Tinfoil. Morse’s policy adds that this depends on the enclave hardware and the attestation working as designed.

E2EE has costs. It needs the Pro plan, it runs only open-weight models such as Qwen, GLM, Kimi and DeepSeek, and it has no web search. The GPT, Claude and Gemini models exist only in Anonymous mode.

Chat history stays on the device. The browser stores chats, images and settings in its own database, encrypted with a data key. That key opens only when two keys are used together: one made in the browser that cannot be copied out, and one Morse keeps and hands over at sign-in. Morse’s policy says plainly that this is not end-to-end encryption against Morse: it holds a key, but it never receives the data the key unlocks. Clearing the browser, losing the device or deleting the account deletes the chats, so Morse offers a backup file, locked with a recovery key or a passkey that Morse never sees. Pro adds encrypted sync across three devices, started with a QR code, where the server only stores changes encrypted under keys the devices hold.

The account is optional for a first try. A message typed on the home page opens a guest chat in Private mode that is not saved. Signing up needs Google or an email address, and the Free plan wants a real email, not a throwaway one. Morse also offers an anonymous account with no email and no name that signs in with a passkey or a saved key, but it needs a paid plan or credits, and nobody can recover it if both the passkey and the key are lost.

Morse still sees some data. Its sign-in provider keeps the IP address and browser of each sign-in, and the human check on the Free plan stores a keyed hash of the IP address. Morse’s own list of limits adds that the relay and the host see which account made a call, the IP address, the time, the model, the size of the request and reply to the nearest kilobyte and the reply’s timing, in every mode. Features around the chat have their own exposure: web search goes to Parallel, which in Anonymous mode may keep the queries, deep research works only in Anonymous and is kept by Parallel, and a plugin vendor sees what Morse sends it.

Trust & Transparency

Watermark Labs publishes six repositories. The enclave that handles every request is under the AGPL-3.0 licence. A second repository lists the enclave images the apps accept, with the date each was pinned or revoked. Two more hold the checking tool and pointers to the public log that records each enclave boot. The last two are the browser exporters that move chats in from other apps. The web app itself, the relay, the accounts and the payments are not published.

The check is built to be run by the reader. The command npx @askmorse/verify confirms that AWS signed each enclave’s attestation, that the enclave’s image is on a list Morse signs and records in Sigstore’s public log, and that a commit in the public enclave repository names that image. Run with --rebuild, it makes a reproducible build of the image on the reader’s machine and compares the result with the one AWS signed.

The repository’s own README lists what the check does not cover. The web app’s check runs in JavaScript that Morse serves, so a different build could skip it, and the README tells users to run the command themselves. The relay picks a chat’s character prompt and could add system text to an encrypted chat without the receipt showing it; binding the character inside the encryption is planned. Models outside the three enclave hosts read requests in plain text. No second person reviews a change to the list of accepted images.

No independent audit or security certification is published. A reader has to rely on the attestation, the public enclave code and the rebuild, not on an outside report. The repositories were created between 19 October and 10 November 2026.

The privacy policy is specific. It says Morse does not train AI on content, sell data or show ads, and uses no advertising trackers or third-party analytics. The server proxy keeps no request logs. What it does count are aggregate statistics with no account attached, and there is no switch to turn that off. If a visitor arrives from one of Morse’s ads on X, Morse tells X whether they signed up or paid, identified only by the click ID on the ad’s link; Global Privacy Control in the browser switches that off. Deleting an account removes the sign-in details and the key to the on-device data. A billing ledger without a name or email is kept, generally for up to seven years.

The company is based in the United States, so US law and legal process apply. Because Morse holds no readable chats, there is little to hand over besides account and billing records and encrypted sync data. No security incident or data breach has been reported. Telemetry is limited to those aggregate counts, with no account attached.

Features

The model list covers chat models from OpenAI, Anthropic, Google, xAI, Meta, DeepSeek, Moonshot AI, Qwen, Z.ai and MiniMax, and image models from Black Forest Labs, ByteDance, Ideogram and Google. Switching between them is available in every mode, but the choice shrinks as privacy rises: Anonymous has the models of OpenAI, Anthropic, Google and Meta, while Private and E2EE are mostly open-weight models.

Around the chat there is web search on every plan in Anonymous and Private chats, deep research that writes a cited report on paid plans, memory of short facts about the user stored on the device, projects with shared files, artifacts for small apps and charts, dictation, and plugins for Notion, Linear, Todoist, Trello and Airtable on Pro. Scheduled prompts allow five tasks on Pro and fifteen on Max. Studio edits images, with upscaling, relighting and changes to one part of a picture. Characters is a role-play section with a community of shared characters, for adults only, with explicit content off by default. Pro unlocks uncensored text and image models. The API is OpenAI-compatible, charged per request, and its TLS connection ends inside the enclave.

Importing is free on every plan: bookmark-style tools run in the browser on another chat app, save a file to the device and open Morse’s import page. Each exporter reads only the site it runs on and sends the user’s sign-in token to that site alone.

Usability

Morse works in a browser. The site lists iPhone and Android apps as coming soon; until they arrive it installs from the browser to a home screen or desktop. A first message needs no sign-up, and a free account uses Google or an email address.

Because chats live in one browser, moving to another device takes a step: a backup file that is restored there, a direct transfer between two signed-in devices, or Pro sync. Sync launched on 10 November 2026. The anonymous account needs payment, and the Free plan needs an email, so staying fully anonymous costs money.

The free allowance is 15 messages and 3 images a day in Anonymous and Private chats, and unlimited character chats. A reply keeps writing when the tab is reloaded or the connection drops.

Price & Value

The free plan is real but capped: 15 messages and 3 images a day in Anonymous and Private chats, plus unlimited chats with characters. It needs a real email address and passes through a human check at sign-up. That is enough for light use.

The paid plans are Pro and Max. Pro costs $19 a month, or $17 a month if the whole $204 is paid for the year. Max costs $49 a month, or $44 a month for $528 a year. The annual price stays the same at renewal. Tax is added at checkout, and no plan shows ads.

What the money buys is worth reading closely. Pro lifts the daily cap on a set of Pro models, among them Claude Sonnet 5.5, Grok 4.7 and an uncensored DeepSeek variant, and it unlocks most of what sets Morse apart: end-to-end encrypted chats, sync, memory, deep research, plugins and uncensored models. The strongest privacy mode therefore sits behind a paid plan. The newest top models are not covered even by Pro and are paid for with credits, 100 credits for $1. Pro includes 250 credits a month, which is $2.50, and Max includes 5,500. Unused plan credits expire at the end of the billing period, while credits bought separately never do. Max suits people who use frontier models heavily. The API has no subscription: credits are added to the account, and each request costs the model’s listed price.

Pros and Cons

  • Three privacy modes chosen per request, and in all three the model company never learns who is asking
  • In E2EE mode the browser checks the hardware attestation of both Morse's enclave and the model company's enclave before encrypting, so Morse cannot read the message
  • Chats are stored on the device, and Morse's servers hold no message they can read
  • A guest chat works with no account and nothing saved
  • An anonymous account with no email or name is available with a passkey or a saved key
  • The enclave code is open under AGPL-3.0, builds can be reproduced, and one command checks what the servers run
  • Morse lists the limits of its own design in the enclave repository
  • No ads, no sale of data, no training on content, no third-party analytics
  • More than 80 models, with search, deep research, memory, image editing, plugins and an API
  • Import of chats from other AI chat apps, free on every plan
  • Encrypted backup file and Pro sync across devices, both locked with keys Morse does not hold
  • End-to-end encryption needs the Pro plan, has no web search, and runs only open-weight models
  • In Anonymous mode the model company reads the message and may keep it briefly, and the zero-retention agreements behind Private mode are not published
  • The web app that encrypts in the browser, the relay, accounts and payments are not open source, and the browser's check runs in code Morse serves
  • No independent audit or security certification has been published
  • Morse holds one of the two keys that open the on-device chats, so they are not hidden from Morse by encryption alone
  • The relay and host see the account, IP address, time, model and message sizes in every mode
  • The relay could add character instructions to an encrypted chat without a receipt showing it
  • The company opened to the public in October 2026, so the track record is short
  • No native iPhone, Android or desktop apps yet
  • Chats exist on one browser unless a backup file or Pro sync moves them, and clearing the browser deletes them
  • An anonymous account requires a paid plan or credits, and the Free plan requires a real email
  • Top models spend credits on top of the plan price

Our Rating

Morse suits someone who wants to pick, message by message, how much the model company learns, and who will pay for Pro to get end-to-end encrypted chats checked by hardware. The design is stronger than a retention promise because a reader can verify the enclave code, and the company says where its own protections end. It suits someone less well if they want the encrypted mode on the free plan, a native app, an outside audit, or a record that goes back further than October 2026. The open enclave code carries the trust, and the closed web app that does the encryption in the browser is the main gap.

  • Privacy Architecture8.7/10

    Model companies never learn who is asking, and E2EE is hardware-verified, but it needs Pro and open models, and Morse holds one of the two keys to the on-device chats.

  • Trust & Transparency6.5/10

    The enclave code is open and checkable, but the web app, relay and accounts are closed, no audit exists, and the company is new.

  • Features8.2/10

    More than 80 models, search, deep research, memory, plugins and an API; E2EE chats have no search and no models from the large labs.

  • Usability6.2/10

    Easy start with a guest chat and import, but no native apps and chats stay in one browser unless moved.

  • Price & Value6.0/10

    Usable free plan, but E2EE and sync need Pro and the top models spend credits on top.

Overall7.4/10

Privacy Architecture 30% · Trust & Transparency 20% · Features 20% · Usability 20% · Price & Value 10%

See our rating methodology →

Ready to try Morse?

Visit Morse →

Last updated: November 14, 2026By Juan Martinez

Missing an app?

Tell us what you could not find. Every suggestion goes onto our list of products to review.

This opens a message in your email app, addressed to hello@privacypick.com. Nothing is sent until you press send there.