Kagi Assistant alternatives: where each one reads your prompt and keeps your threads
The Morse teamPublished Facts checked 12 min read
The Kagi Assistant alternatives that keep your chat history out of a readable server database, as of November 2026, are Morse, Brave Leo, Duck.ai, Proton Lumo, and Venice, while Perplexity, like Kagi, saves your threads on its own servers. Kagi Assistant saves each thread on Kagi's servers and passes every message on to the model's provider, so when you switch, check where your prompt gets decrypted and where your history ends up.
We read Kagi's Assistant docs, its LLMs & Privacy table, its pricing page, its privacy policy (last changed September 22, 2026), its Privacy Pass docs, and its changelog on November 16, 2026, then each alternative's own pricing and privacy pages the same day. Kagi gets a lot right. It shows no ads and says it "is fully funded by its users," it publishes how long each model provider keeps an Assistant request, and it credits you a full month for any month you don't use your plan (Kagi).
What are the best Kagi Assistant alternatives?
As of November 16, 2026. Every cell comes from the company's own pages, linked in the sections below.
| App | Where your prompt is decrypted | Who checks it | Where history lives | Trains on your chats | Ads | Models | Paid from |
|---|---|---|---|---|---|---|---|
| Kagi Assistant | Kagi's servers, then the model's provider | Kagi's contracts with providers | Kagi's servers, 24 hours by default, up to permanent | No, by Kagi or its providers | None | Claude, GPT, Gemini, Grok, and open models; flagship ones on Ultimate | Starter $5 a month; Ultimate $25 |
| Morse | Anonymous: the provider, without knowing it's you. Private: a host that agreed to keep nothing. E2EE: only an enclave your browser verified | E2EE: your browser checks the enclave's attestation | In your browser, encrypted | Morse doesn't train AI on your chats | None | 80+, including GPT, Claude, Gemini, Grok, DeepSeek, GLM, and Kimi | Free; Pro $19 a month |
| Brave Leo | Leo's servers, which keep no prompts; GLM 5.3 Flash only inside an attested enclave | For GLM 5.3 Flash, Brave's gateway server, on your browser's behalf | On your device | No | None described for Leo | Open models and Claude | Free; Premium $14.99 a month |
| Duck.ai | The model's provider, reached through DuckDuckGo with your IP address removed; two free models inside an enclave | DuckDuckGo's agreements; for two models, the enclave | On your device, or encrypted Sync & Backup | No, by DuckDuckGo or the providers | DuckDuckGo earns from search ads | GPT, Claude, Mistral, and open models | Free; Plus $9.99 a month, bundled with a VPN |
| Proton Lumo | Proton's own GPU servers | Proton's word, since it runs the servers | On your device, synced with zero-access encryption | No | Messages never "used for targeted ads" | Qwen 3.5, GLM 5.3, and Apertus 1.5 | Free; Plus $12.99 a month |
| Venice | Anonymous: the provider. Private: Venice or a zero-retention partner. TEE and E2EE: an enclave | Remote attestation, for TEE and E2EE | On your device | No: its terms bar Venice and its providers from training on your content | No ads in chats described; its privacy policy lets advertising partners use cookies | GPT, Claude, Gemini, Grok, and open models | Free; Pro $18 a month |
| Perplexity | Perplexity's servers, then the model's provider | Perplexity's own policies | Perplexity's servers while your account is open | Yes by default, until you opt out | Shares data with third parties to measure its own advertising | Top models; frontier ones on Max | Pro $17 a month, billed annually |
Why do Kagi subscribers look for an alternative?
Kagi's own pages point to four.
Your history sits on Kagi's side. Assistant threads live in Kagi's database under your account; by default a thread expires 24 hours after its last message, and you can make it permanent (Kagi), and Kagi's November 5, 2026 changelog redid the Assistant's icons because users found it "too hard to tell when a temporary thread was actually temporary" (Kagi). Who can read them, and for how long, is in is Kagi Assistant private.
The model maker still gets the words. Kagi strips identifiers, but its provider table shows Claude, GPT, Grok, and Mistral requests held for weeks, and it now lists Gemini 4 Carbon at "up to 30 days" through Google Vertex (Kagi).
Privacy Pass stops at Search. The privacy policy counts Privacy Pass among its routes to "provable anonymity guarantees" (Kagi), yet the Privacy Pass docs still list the Assistant as "soon to follow" (Kagi), so Assistant use stays tied to your login.
The best models cost $25 a month. "Premium models are only available in our Ultimate plan," and AI use is capped by value: "a $25 monthly plan provides up to $25 worth of token usage across all models," of which $5 is a 20% markup for search and infrastructure. Go over and "new AI interactions will be disabled" until the next billing cycle, unless you upgrade or renew early (Kagi). Kagi says your data is "never used to train AI models (not by us or by the LLM providers)." Starter, at $5 a month for 300 searches, and Professional, at $10, get the Assistant's Quick mode and the models marked "All," while Ultimate adds Research mode and flagship models such as Claude Opus 5.5, GPT-6.1 Sol, Gemini 4 Argon, and Grok 4.7 (Kagi). Kagi's trial covers 100 Assistant interactions.
What do you give up by leaving Kagi Assistant?
Mostly the search engine under it. The Assistant's web access is "powered by Kagi Search," and your Lenses and Personalized Results carry into its searches. It also takes uploads up to 30 MB, fetches web pages and documents up to 50 MB, has voice input and custom assistants, and can share a thread by link (Kagi).
You don't have to give up Kagi Search to move your chats. Since November 5, 2026, anyone can run 50 searches on kagi.com without an account (Kagi), and Professional keeps unlimited searches at $10 a month. Before you cancel the AI side, export what you want to keep: each thread's menu exports it as Markdown or JSON, and Kagi notes that "any text editor" opens the Markdown file (Kagi).
Is there a Kagi Ultimate alternative with the same models?
Yes, for several of them, and some sit on other apps' free plans. Kagi's model table puts GPT-6 Luna on Ultimate only, while Morse and Duck.ai both offer it free. Here is Kagi's plan for each model next to the plans Morse and Duck.ai list for it, as of November 16, 2026.
| Model | Kagi plan | Morse plan | Duck.ai plan |
|---|---|---|---|
| GPT-6 Luna | Ultimate, $25 | Free | Free |
| Claude Haiku 5.5 | Every plan | Free | Free |
| GLM 5.3 Flash | Every plan | Free | Not listed |
| gpt-oss-120b | Every plan | Not listed | Free, in an enclave |
| DeepSeek V4.1 Flash | Every plan | Pro, $19 | Not listed |
| Claude Sonnet 5.5 | Ultimate, $25 | Pro, $19 | Plus, $9.99, bundled |
| Grok 4.7 | Ultimate, $25 | Pro, $19 | Not listed |
"Not listed" means the model isn't named on the plan or model pages we read; check the app's model menu before you decide. Sources: Kagi, Morse's plans, and DuckDuckGo. Brave Leo also offers GLM 5.3 Flash free, end-to-end encrypted (Brave).
1. Morse: no thread on our servers
Kagi leaves two copies of a conversation: the thread in its database and the message at the model's provider. Morse removes the first by design and lets you choose what happens to the second. Your chat history is stored in your browser, encrypted, not on Morse's servers, and those servers don't keep messages or log what you write. There's no Thread Saving choice to make before you start and no 24-hour timer to trust. Morse doesn't train AI on your chats, sell your data, or show ads (Morse privacy).
You get 80+ models, open and closed, in one account. The Free plan has 15 messages a day with GPT-6 Luna, Claude Haiku 5.5, GLM 5.3 Flash, and more, plus web search, with no card. Pro, $19 a month or $17 billed yearly, adds unlimited Pro models such as DeepSeek V4.1 Flash, Claude Sonnet 5.5, and Grok 4.7, unfiltered models, end-to-end encrypted chats, memory across chats, and deep research. Max is $49 a month. Each chat gets its own privacy mode, which decides who reads your prompt:
- Private, for the questions you'd send to Kagi's "Not stored" models. Private is the default for a new chat and runs open-weight models like DeepSeek, Kimi, and GLM, plus Grok, which Kagi's table gives 30 days at xAI. Morse has zero data retention agreements with its inference, search, and other providers, so they've agreed to keep nothing. Where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't. What zero data retention means covers the promise.
- Anonymous, for GPT, Claude, and Gemini with your identity stripped, as Kagi's proxy does. Our proxy removes your account, email, and IP address and sends the message under Morse's own account. The provider sees the message 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. Deep research runs in Anonymous mode for now.
- E2EE, for questions that don't need the web. Your browser checks the enclave's hardware attestation, then seals the message so only that enclave can read it, and Morse's servers relay ciphertext they can't read. E2EE comes with Pro, runs open-weight models like GLM, Kimi, and DeepSeek, and has no web search or image generation, because its models can use only tools that run on your device. The guarantee rests on the enclave hardware and attestation working as designed. End-to-end encrypted AI chat walks through the check.
Since your history lives in your browser, clearing the browser, losing the device, or deleting your account deletes those chats, and Morse can't recover them, so copy out any answer you'll need again. 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.
2. Brave Leo
Leo is the private AI search assistant built into the Brave browser, on desktop and mobile, and it needs "no account or login." It "uses Brave Search to enhance answers," and keeps chat history "locally on your device" (Brave). Brave's privacy notice, last updated September 23, 2026, says it doesn't "store or retain prompts, responses, context, or personal data on our servers," doesn't collect identifiers such as your IP address, and doesn't use conversations for training (Brave).
Brave deserves credit for pairing encryption with search. On October 15, 2026, it made Leo's GLM 5.3 Flash chats end-to-end encrypted. Brave's AI Gateway, on Brave's server, verifies the enclave's attestation on your browser's behalf, and Brave says "today, your browser relies on the gateway for this check." Only after it passes does your browser get the key to seal the prompt with Oblivious HTTP. The model can also "call a search tool directly from inside the TEE," so an encrypted chat can still search the web (Brave). Brave's relay sees your IP address while it passes the sealed message along.
Brave's Leo page says its versions offer "the latest models from Mixtral, Anthropic's Claude, and Meta's Llama" (Brave). Leo lives only inside Brave's browser. A thumbs-up or thumbs-down sends Brave "the full text of the current conversation," deleted after a year (Brave). Premium is $14.99 a month or $149.99 a year for more models and higher limits (Brave). We compare it in detail in Morse vs Brave Leo.
3. Duck.ai
Duck.ai is free with no account. Its free models are Claude Haiku 5.5, Mistral Small 4, GPT-6 Luna, GPT-5.4 mini, gpt-oss-120b, and Gemma 4 31B (DuckDuckGo). DuckDuckGo removes "all metadata that contains personal information (for example, your IP address)" before a prompt reaches the provider, and its agreements bar providers from using prompts to improve their models and require deletion "at most within 30 days, with limited exceptions for safety and legal compliance." Its privacy terms were last updated on November 16, 2026, to clarify supported regions (DuckDuckGo).
Two of the free models, gpt-oss-120b and Gemma 4 31B, run in a Trusted Execution Environment and are labeled "zero provider visibility." Uploaded images and files still go to an outside moderation service "outside of the encrypted channel" (DuckDuckGo). Recent chats are saved on your device, or in Sync & Backup, where "the decryption key is only stored on your devices" and chats left untouched for 18 months are deleted. Plus, $9.99 a month, adds GPT-5.6 Terra and Claude Sonnet 5.5, and Pro, $19.99, adds Claude Opus 5.5 and GPT-5.6 Sol, both inside a subscription "including VPN and advanced AI" (DuckDuckGo). For search, its Search Assist answers are "sourced from the web" on the results page (DuckDuckGo). DuckDuckGo says it makes money "from privacy-respecting search ads" (DuckDuckGo). More in Duck.ai alternatives.
4. Proton Lumo
Lumo trades the big closed models for servers Proton runs itself. Its models, "Qwen 3.5, GLM 5.3, Image-Turbo, FireRed-Image-Edit-1.1, and Apertus 1.5," "run exclusively on servers Proton controls," web search is optional, and your history is "stored on your device and synced to Proton servers using zero-access encryption" (Proton). Prompts are encrypted so "only the Lumo GPU servers are able to decrypt them," and Proton's security write-up says plainly, "we need to provide the LLM with cleartext messages," which it calls user-to-Lumo (U2L) encryption rather than end-to-end (Proton). Proton also says "Lumo doesn't use your chats to train the AI models," except anonymized feedback on Apertus 1.5 that you choose to send (Proton).
The free plan includes "basic chat history, web search," and small uploads. Plus is $12.99 a month, or $9.99 a month billed at $119.88 a year, and Proton says your messages are never "used for targeted ads" (Proton). There's no GPT or Claude. See Lumo alternatives.
5. Venice
Venice sorts every model into four privacy modes, and its privacy page is frank about the weakest. For GPT, Claude, and Gemini, "your identity is obscured from the provider, but you should assume the provider is storing your content." Private runs on Venice's GPUs or zero-retention partners, offers a "wide model selection (including Grok)," and is "trust-based." TEE models are "verified by remote attestation," TEE and E2EE models need Pro, and E2EE comes with "no web search or memory." In every mode, "your conversation history stays on your device" (Venice). Web search can use "Brave with zero data retention and anonymized Google search" (Venice). Free allows 10 text prompts a day, and Pro is $18 a month or $180 a year (Venice). Venice's terms say it "shall not use, and shall require that third-party large language model providers... do not use, any User Content to train" models (Venice). Its pricing and privacy pages describe no ads in chats, though its privacy policy says "advertising" partners may use cookies and similar tools on the service (Venice). More in Venice AI alternatives.
6. Perplexity
Perplexity has the most research tools and keeps the most. Pro is $17 a month billed annually, with deep research, model choice, and its Computer agent, and Max, $167 a month billed annually, adds frontier AI models and the highest limits (Perplexity). Its privacy notice, last updated October 23, 2026, says it "may share data with third parties to measure the performance of our advertising," for example through "a cookie or a pixel" (Perplexity). Its help center, as archived on November 10, 2026, says "AI Data Retention is enabled by default" for Free, Pro, and Max, and that data may train its models "unless you opt out through Account Settings" (Perplexity). It keeps personal information "as long as your account is active" (Perplexity). Kagi vs Perplexity sets the two side by side.
How we compared them
We read every page linked above on November 16, 2026, and took each company at its written word; we ran no tests. Perplexity's site refused our requests, so we read its Pro page through a reader service and its two help articles from Internet Archive copies made November 10 and November 12, 2026. Where a company's pages don't answer a column, the table says so.
Should you keep Kagi or switch to Morse?
- Choose Kagi if you want AI on top of a search engine you pay for, with your Lenses and personalized results shaping the answers, flagship models in one $25 plan, and you're fine with threads on Kagi's servers for a day or longer.
- Choose Morse if you want GPT, Claude, Gemini, Grok, and open-weight models in one account with no thread on our servers, Private chats whose hosts agreed to keep nothing, and end-to-end encrypted chats your own browser checks.
Ask the next question you'd have typed into Kagi on the Morse home page without an account, or start free with no card; Pro adds E2EE chats at $19 a month.
Morse publishes this comparison. We've tried to keep it fair; if something's wrong or out of date, email hello@trymorse.ai.
Kagi is a trademark of Kagi Inc. Morse is not affiliated with or endorsed by Kagi Inc.
Common questions
Is there a free Kagi Assistant alternative?
Yes. As of November 16, 2026, Duck.ai and Brave Leo are free with no account, Lumo's free plan includes web search, Venice Free allows 10 text prompts a day, and Morse's Free plan has 15 messages a day with web search. Kagi's own trial covers 100 Assistant interactions.
Can I move my Kagi Assistant threads to another app?
Kagi lets you export a thread as Markdown or JSON from the menu beside it, and any text editor opens the Markdown file. Export before you cancel, because a thread on the default setting expires 24 hours after its last message.
Which Kagi Assistant alternative has end-to-end encryption?
Morse (E2EE chats on Pro), Brave Leo (its GLM 5.3 Flash model), and Venice (E2EE models on Pro) encrypt a prompt on your device so only a verified enclave can read it. Duck.ai runs two free models in a Trusted Execution Environment, and Proton says Lumo has to "provide the LLM with cleartext messages" on servers Proton controls.
What does Kagi Ultimate add over Professional?
Research mode and Kagi's flagship models, such as Claude Opus 5.5, GPT-6.1 Sol, Gemini 4 Argon, and Grok 4.7, for $25 a month instead of $10. AI use is capped by value: a $25 monthly plan covers up to $25 of token usage, which includes a 20% markup.