Private intelligence · in active development

Searxly AI.

A private intelligence layer for the browser. Run it fully on your device or on independent, decentralized infrastructure — built on open models, grounded in your own private search, and never trained on your chats. The intelligence is open. The privacy is ours.

On-device or private cloud · Open models · Your data, your rules
The approach

We didn't build another black box.

Most browser AI asks you to trust one company's opaque model and its data practices at the same time. Searxly AI separates the two. We run leading open models — auditable, swappable, no single-vendor lock-in — and wrap them in a privacy and grounding architecture that we own and keep: your search stays private, your data stays yours, and you choose where the compute happens.

Where it runs

You decide where the thinking happens.

No hyperscaler quietly harvesting your prompts. The safe choice is the default choice.

On your device

Apple Intelligence and local models run entirely on your Mac. Prompts, context and your private data never leave the machine — verifiable with a network monitor.

Independent cloud

When you want more horsepower, the private cloud runs open models on a decentralized GPU network — independent compute, not a Big-Tech data center built to profile you.

One honest switch

A single toggle in Settings — nothing more to configure. Local stays the default; the private cloud is opt-in, and what it does is stated in plain language.

What makes it Searxly

The model is a part. The product is the architecture.

A model on its own is a commodity. What you actually feel is how it's grounded, what it remembers, what it's allowed to do — and what it refuses to leak.

Grounded in your private search

Answers are synthesized from your own local SearXNG instance — the same private engine that powers Searxly search — with inline citations. Never a public index, never an ad network, never your intent for sale.

Private memory over your own data

Optional semantic recall across your history and bookmarks, computed on-device so the AI can answer "what was that article I read about…" without a byte of it ever being uploaded.

Agentic — but only with consent

It can search privately, open the right official site (resolved from a curated, auditable entity database — not a lucky guess), and read files you attach. Every action is visible, logged, and reversible. You can require a confirmation for anything that touches the network.

Open models, no lock-in

Built on leading open-weight models you can inspect and we can swap as the field moves. You're never trapped behind one vendor's black box, one vendor's terms, or one vendor's outage.

Privacy by architecture

No training on your conversations. No ad identity, no behavioral profile. A master kill-switch makes the whole layer inert, panic-wipe clears its memory, and every model call is written to an activity log you can read and export.

Typical browser AI

  • Your prompts train someone else's model by default
  • Silent profiling tied to an ad identity
  • Answers pulled from a tracking-funded index
  • One opaque vendor, one set of terms, no way out
  • Runs only where they say, on their hardware

Searxly AI

  • No training on your conversations
  • No ad identity, no behavioral profile
  • Grounded in your own private SearXNG, with citations
  • Open models — auditable and swappable
  • On-device or independent, decentralized cloud — your call

“Capability without surveillance.”

That's the whole bet. If an AI can only be useful by watching you, it isn't the one we want to ship — and it isn't the one you deserve. So we made the intelligence open, the compute your choice, and the privacy non-negotiable.

Searxly Agent

An assistant that can actually do things — with your say-so.

Chatting is the easy part. The harder, more useful problem is letting the model act inside the browser — reliably, and only with consent. That's the Searxly Agent: tool calling wired into the browser and grounded in your private data, never an open door.

Working today

Private web search

The agent searches through your own SearXNG, reads the results, and answers in chat with citations — never a public index or an ad network. Trigger it yourself with the "Web search" chip, or let the model decide.

Open the right site

Ask it to open something and it resolves the real, official destination from a curated, auditable entity database — instantly and correctly, instead of guessing from a search result. Explicit navigation only; it won't wander off.

Typed, native tool calling

On supported Macs it uses the model's native, structured tool calling — not brittle text parsing — so the model decides when to act and the browser runs the bound action. A reliable marker-based path covers everything else.

Consent · logged · reversible

One toggle decides whether tools run automatically or ask first. Every tool call is shown in the transcript and written to an activity log you can read and export — nothing happens behind your back.

What we're building

Better tool calling, more of the browser.

Tools for your own data

Search your history and bookmarks in plain language, open a set of results in tabs, bookmark a page with a note from the conversation, or spin up a fresh private search tab — all on-device, all from a single sentence.

Reliable multi-step flows

Stronger grounding, classification and confirmation so the agent can chain a couple of steps — "look this up, then open the official site" — without misfiring, even on small on-device models.

System-level reach

Exposing the agent's actions to Spotlight, Shortcuts and Siri, so "private search for X" or "start a Searxly chat" become first-class system commands — without giving up a shred of the privacy contract.

Early access

Be first to try Searxly AI.

On-device chat is already taking shape inside the app, with the private cloud rolling out next. Early builds reach the community first — follow along for the announcement and grab the app early through the holder waitlist.