Is Kagi Assistant private? Where your threads live and who can read them

The Morse teamPublished October 30, 202610 min read

As of October 30, 2026, Kagi Assistant is private from advertisers and from AI training, but not from Kagi: your threads are saved on Kagi's servers under your paid account, and each message goes to the model's provider, which keeps it for 30 days in the case of Claude, GPT, and Grok. By default, Kagi deletes a thread 24 hours after its last message, and Privacy Pass, the token system that lets you search without your account, doesn't yet work in the Assistant.

We read Kagi's privacy policy (last changed September 22, 2026), its privacy rights addendum, its Assistant and LLMs & Privacy help pages, its Privacy Pass docs, its pricing page, its changelog, and its answers on its own feedback forum, all on October 30, 2026. Every fact below links to the Kagi page it came from. Some of this is to Kagi's credit. It publishes how long each model provider keeps an Assistant request, lists each of its debugging logs with the number of days it lasts, and keeps a warrant canary in its privacy policy.

Kagi Assistant privacy at a glance

Does Kagi Assistant train on your data?

No, according to Kagi, though its own table leaves one model unverified. Kagi's Assistant page says "your data is never used to train AI models (not by us or by the LLM providers)" (Kagi). When Kagi opened the Assistant to every plan on April 17, 2025, it said that applies to third-party providers "under strict contractual terms" (Kagi). There's no training switch to find.

Kagi's own LLMs & Privacy page marks every provider "No" for training on API data except one. For Qwen3.8 Flash, a model open to every plan, both the training and the retention columns read "Not verified" (Kagi). Meta's Muse Spark 1.3 is marked as not training, but its retention reads "Not specified," because "Meta does not state a retention period for API requests." If you want Kagi's no-training promise to hold for every message, pick another model until Kagi updates that row.

How long does Kagi Assistant keep your threads?

A day by default, or as long as you choose. The Assistant help page says threads are kept "for 24 hours after the last message," and that Thread Saving in the Assistant's settings offers "temporary (threads expire 24 hours after the last message) or permanent" (Kagi). Kagi's August 21, 2026 changelog added two more lengths: "Now you can choose whether temporary threads stick around for 24h, 7 or 30 days" (Kagi). The help page still listed only the 24-hour option on October 30, 2026.

Three details change what that setting does:

• It's fixed when a thread starts. Kagi says "the thread-saving setting is applied when the thread is created," so changing it doesn't shorten threads you already saved (Kagi). A single thread can be made permanent from its menu.
• Temporary threads once outlived their timer. Kagi's July 30, 2026 changelog lists "Temporary threads don't actually get removed" among the bugs it fixed (Kagi).
• Reporting a reply keeps a copy. Since July 30, 2026, a thumbs-down opens a report form, and "the full thread is shared with Kagi for review." Kagi's privacy policy says reported threads are "retained for 30 days and then automatically deleted" (Kagi).

Kagi's general FAQ still carries an older answer: "during beta, Kagi will store conversations for 7 days in order to handle bug reports," with a plan to cut that to 24 hours at launch (Kagi). The privacy policy is the current word. It says threads are deleted "after one day" by default, and "after this period, they are permanently removed" (Kagi). To clear everything at once, the July 30 changelog says you can "export all your threads or permanently delete them at once from Settings > General."

Can Kagi read your Assistant threads?

Its systems and some of its engineers can. A user asked on Kagi's feedback forum in September 2024 how threads are stored, and Kagi answered that it uses "a standard hosted DB," where "data is encrypted at rest and in flight but is available as plaintext to authenticated users (e.g. the assistant application, and operations engineers with monitored access)" (Kagi Feedback). The same reply said stronger encryption would cost features like "smart titles" and "thread search." We found nothing newer on Kagi's pages that changes this.

Unlike your searches, threads belong to your account. Kagi's privacy policy says requests to its AI tools "are not saved or associated with a customer account," but opens that sentence with "Unless stated below," and the Assistant's thread rules follow (Kagi). Kagi's iOS and Android apps, launched July 30, 2026, promise that "your threads and Custom Assistants stay with you" from one device to the next (Kagi).

Anyone you give a link can read a thread too: the Assistant lets you "share threads with others using a link" (Kagi). The August 21, 2026 changelog lists a fix to keep shared threads out of search engine indexes (Kagi).

What do the AI companies see?

Your message, without your Kagi account. Kagi says "no account information is shared with the LLM providers" (Kagi), and its privacy policy says that for the Assistant it proxies "all connections and data through our servers" (Kagi). For OpenAI's models, Kagi adds that "the Assistant does not include a unique user identifier for these requests" (Kagi).

The provider still reads the text and may keep it. Kagi's table lists 30 days for Claude, GPT, Grok, and Mistral's models, 24 hours for Gemini, and "Not stored" for open-weight models such as DeepSeek, Kimi, and GLM. Claude prompts that Anthropic flags are kept for 2 years. Files travel with your messages: the Assistant takes uploads up to 30 MB, including PDFs, spreadsheets, images, and audio (Kagi). We set Kagi's provider table beside other multi-model apps in T3 Chat alternatives.

Does Privacy Pass work with Kagi Assistant?

Not yet. Privacy Pass swaps your login for cryptographic tokens, so "your searches can't be linked back to your account or to each other," and it needs the Professional, Ultimate, Family, or Team plan (Kagi). The docs say "at launch, Privacy Pass will only be used to authenticate Kagi Search," and put Kagi Assistant first on the list of products to follow.

A user asked for it on Kagi's forum in October 2025, noting that using Privacy Pass on search "currently disables the assistant function," and Kagi tagged the request Planned on December 5, 2025 (Kagi Feedback). Kagi's October 16, 2026 changelog says "Privacy Pass now works with Quick Answer & Inline Summarize features," two AI features inside Search, and doesn't mention the Assistant (Kagi).

Until then, Kagi's other options loosen the tie between the account and you. Its privacy policy lists "a pseudonymous email address," "paying for their plan using cryptocurrency," and access over Tor (Kagi), and its privacy page says "You do not need to use an email address that can easily identify you" (Kagi).

Is Kagi Assistant more private than using ChatGPT or Claude directly?

Kagi's FAQ says yes, "because Kagi acts as a proxy between you and the LLM (both from an internet connection and payments perspective)," and because with API access "the provider does not use the data for training" (Kagi). By Kagi's own pages that holds up: the model's maker gets your message without your account, your connection, or your card. The difference is that Kagi now holds the thread. How the big apps treat chats under their own defaults is in is ChatGPT private and is Claude private.

Can Kagi hand your threads to the police?

Kagi's privacy policy doesn't say how it answers legal requests, but it does carry a warrant canary. As of October 30, 2026, it says Kagi has received "0 National Security letters," "0 Gag orders," and "0 Warrants from any government organization," and that customers' private information "has never been disclosed or seized" (Kagi). Kagi became a public benefit corporation in early 2024 (Kagi), and its terms are governed by Delaware law. A saved thread is a record Kagi holds, the kind of data a subpoena or warrant can reach, and we cover how in can AI chats be subpoenaed. This isn't legal advice.

Which Kagi plans include the Assistant?

All of them, including the free trial. Kagi's help page says "Assistant is available to all members," with premium models only on Ultimate (Kagi). The trial covers 100 searches and 100 Assistant interactions. Starter is $5 a month and Professional $10, both with the Assistant's Quick mode, and Ultimate, $25, adds Research mode and flagship models such as Claude 5.5 Opus, GPT-6.1 Sol, Gemini 4 Argon, and Grok 4.7 (Kagi). Use is capped by value: "a $25 monthly plan provides up to $25 worth of token usage," with a 20% margin built in (Kagi). Kagi shows no ads, and its pricing page says it "is fully funded by its users."

How Kagi Assistant's privacy changed, 2025 to 2026

How Morse keeps your threads out of a server database

Kagi strips your account from every message, and two copies remain. The thread sits in Kagi's database, where, by Kagi's 2024 answer, its application and operations engineers with monitored access can read it, and the message itself goes to the model's provider. Privacy Pass can't yet separate the first from your account, and no Kagi setting reaches the second.

With Morse, the first copy never exists. You get over 30 models, open and closed, under one account, and no thread is saved on our side. Your chat history is stored in your browser, encrypted, not on Morse's servers, and our servers don't keep messages or log what you write. There's no thread timer to trust, since a temporary chat is kept in memory only. Morse doesn't train AI on your content, sell your data, or show ads (privacy policy).

For the second copy, pick a privacy mode for each chat:

• E2EE, for the threads you'd least want an operations engineer or a model host to see. Before a word leaves your device, your browser verifies the secure enclave's hardware attestation and seals the message for that enclave alone. What crosses Morse's servers is ciphertext they can't read. It runs only open-weight models like GLM, Kimi, and DeepSeek, with no web search or image generation, and its guarantee depends on the enclave hardware and attestation working as designed. End-to-end encrypted AI chat walks through the check.
• Private, for Grok and open-weight models without the 30-day copy Kagi's table lists for Grok. Private is the default for a new chat. Morse has zero data retention agreements with its inference, search, and other providers, so they keep nothing. Where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't.
• Anonymous, for GPT, Claude, and Gemini without your identity. Like Kagi, our proxy strips your account, email, and IP address, and sends the message under Morse's own account. The provider sees it without knowing it's you, and may keep it for a limited time to check for abuse, under its own policy, so leave names out of anything sensitive.

Kagi deletes a thread on its own servers when the timer runs out. On Morse, the copy in your browser is the only copy of your history: clear the browser, lose the device, or delete your account, and those chats are gone, with no way for Morse to bring them back. Copy out anything you want to keep. Because we don't store chats, we usually have very little to give authorities. You sign in with a code sent to your email, or with Google.

Kagi Assistant settings to check, and the Morse answer

• Threads you'd rather not keep: in Kagi, set Thread Saving to temporary at 24 hours before you start a thread, and clear old threads from Settings > General. On Morse, history is stored only in your browser, encrypted.
• Training: in Kagi, avoid Qwen3.8 Flash while its row reads "Not verified." Morse doesn't train AI on your content.
• A thread tied to your account: in Kagi, sign up with an email that doesn't name you; Privacy Pass doesn't cover the Assistant yet. On Morse, there's no thread on our servers to tie to anything.
• Copies at the provider: in Kagi, choose open-weight models, which its table marks "Not stored." On Morse, use Private or E2EE.
• Reports: in Kagi, don't thumbs-down a sensitive thread, since the whole thread goes to review for 30 days.

Ask the question you'd have typed into Kagi Assistant on the Morse home page, no account needed. To keep your history, create a free account without a card, and add Pro, $19 a month, when you want E2EE threads.