How NewEyes AI Uses TinyFish for the Page Behind a Picture

At a glance

• Company: Collov Labs, creator of NewEyes AI, a visual AI agent that turns what the camera sees into finished tasks. It runs on phones, with early access on Ray-Ban Meta glasses.
• Use case: Live-web answers to the questions a picture raises but cannot settle on its own.
• The stack: NewEyes draws on several web tools and providers. TinyFish is one of them, responsible for three well-defined jobs.
• Where TinyFish helps: Live official pages read in a real browser, including pages a static fetch misses; recent news within a set time window; and research papers as structured records.
• TinyFish products: Search and Fetch.
• In short: When the answer lies outside the frame, NewEyes turns to TinyFish for three things: the official page as it stands now, dated coverage and the research itself. Everything else stays with NewEyes' other web tools.

Company overview

NewEyes AI is personal visual intelligence from Collov Labs. Point a phone or glasses at something, say a single line, and the agent carries the task forward: it logs the meal, styles the outfit, compares the prices. NewEyes launched publicly in September 2026 and is available on iOS and Android, with early access on Ray-Ban Meta glasses.

The problem: what the camera cannot see

A photo of the Ambassador Bridge shows the span, not the wait at the booth. An older photo of Hawaiʻi's Hōlei Sea Arch shows a landmark that, since late September 2026, no longer stands. Nothing in either frame can say so. Nor do people spell out what they need. Instead of a search query, they make an offhand remark, and it falls to NewEyes to decide which part of the live web holds the answer:

• The official word, right now. Live status, such as the posted wait at a border crossing, kept on a single page that changes through the day.
• What has happened since. The latest coverage, bounded by a time window and dated story by story.
• The research itself. The published paper behind a slide, even one from this year.

For all three, NewEyes now calls on TinyFish, alongside the web tools it already runs.

Why Collov Labs added TinyFish

A lookup matters most while the user is still looking, and a single remark can call for several at once. Speed is the quality Collov Labs singles out in TinyFish: it is fast enough that NewEyes can afford a second or third lookup whenever a moment calls for one.

“TinyFish's latency is really good. It's fast, which makes it a great fit for running things in parallel and at scale.”
Yijia Zhou, Research Scientist, Collov Labs

Where TinyFish fits

Inside NewEyes, TinyFish handles three jobs.

• Read the page. Some answers have one authoritative home: the official page that publishes them. TinyFish Fetch opens that page in a real browser and returns clean content. Shown a photo of the Ambassador Bridge between Detroit and Windsor, NewEyes reads the crossing's posted wait times directly from the official border wait-times page.
• Search the news. TinyFish Search returns stories from a set time window, each with its publisher, its date and usually a thumbnail. Shown the older Hōlei Sea Arch photo, NewEyes reports that the arch collapsed into the Pacific in late September 2026 and lists the coverage, each story dated.
• Find the paper. TinyFish Search also returns research, including work published this year, as structured records: authors, year, venue, citations and, when there is one, a link to read the paper. From a slide summarizing a study, NewEyes identifies the published CHI 2026 paper.

Results

• Answers from the source. Live status as the official page reports it at the moment of asking.
• News with provenance. Coverage within a set time window, every story dated and attributed.
• Research on the record. A study on a slide, traced to its published paper.
• Fast to adopt. Pages arrive as clean content and search results as structured records, so both fold straight into NewEyes' answers. Integrated in an afternoon.

What's next

NewEyes is built to finish tasks, not only to answer questions. Collov Labs is exploring TinyFish Web Agent for that next step: once the assistant understands what it is looking at, it could act on the relevant web page with the user's approval. The team is also exploring TinyFish deep research, now in beta, for questions that merit a cited report.

Read Collov Labs' side of the story: “Giving a Camera-First Assistant a Web Arm” (coming soon)

Building an assistant that needs to read the live web?

Talk to Us | Explore the Playground | Read the docs