How Tethral Reads Grocery Data Trapped in Images with TinyFish
At a Glance
- Company: Tethral is a governed agent platform for everyday life, making IoT physical and agentic
- Use case: Reading weekly grocery circulars and turning them into structured, verbatim price data
- Previously: Firecrawl, Perplexity API, SerpAPI, and in-house scrapers
- Now: TinyFish Search, Fetch, and Web Agent
- Why TinyFish: One web layer for discovery, extraction, and browser navigation across grocery sites that behave very differently
Company Overview
Tethral is building a governed agent platform for everyday life. A person says, in plain language, what they want handled, and the system turns that intent into a standing routine that can run safely on their behalf, with receipts for what it did.
They're starting with grocery shopping. A user might tell Tethral: watch for beef going on sale at my store, match what's discounted against the meals my family cooks, and every Friday send me a plan with real prices and what dinner will cost for the five of us.
That one instruction becomes a recurring agent workflow. Tethral resolves the store, reads that week's sale data, matches it against the household's meals, and can surface something as specific as:
"80% Lean Ground Beef, $3.59/lb, was $4.65 (23% off), best price in 8 weeks."
Every number is traceable to the page it came from.
Their goal is to make grocery savings useful without turning couponing into a part-time job, especially for people for whom chasing weekly circulars is genuinely difficult: older adults aging in place, people with accessibility needs, and busy caregivers.
The problem
Before Tethral can reason about what's on sale, it has to reliably read what's on sale.
And grocery data is surprisingly difficult to access.
Weekly circulars are one of the last major consumer surfaces where important information is still trapped inside images and interactive, store-gated widgets.
Prices aren't always published as clean text. Some retailers only reveal their circular after a user selects a specific store. Others require JavaScript to render the page correctly.
That means the data effectively falls into two classes: retailers with sale pages that can be fetched directly, and retailers where the content only appears after interacting with the site in a running browser.
For Tethral, that makes the product a web data problem before it becomes an agent problem.
If the agent can't reliably read the circular, there's no routine to run.
Before TinyFish
The team was using Firecrawl, Perplexity API, SerpAPI, and its own scrapers to piece together the web layer.
Each solved part of the problem, but managing them together created another problem.
Different providers meant different accounts, different integrations, inconsistent quality, and multiple places where something could break.
When a grocery price failed to make it from a retailer's website into the agent workflow, figuring out where the failure happened became part of the job.
Tethral wanted a simpler web layer that could handle the different ways grocery data appears online without stitching together separate infrastructure for every step.
Why TinyFish
Reading a grocery circular requires more than one kind of interaction with the web.
First, the system has to identify the right store and find its current deals page. Then it has to extract the actual sale information from that page. And when the circular sits behind an interactive store selector or browser-dependent interface, it needs to navigate the site before that information even becomes available.
TinyFish gave Tethral those capabilities through Search, Fetch, and Web Agent.
Instead of sourcing and troubleshooting separate providers for each part of the workflow, the team could build against one web layer.
"If you're using 5 tools for 1 job it's not just 5 tools, it's 5 accounts, 5 logins, 5 troubleshooting connections and 5x the time."
- John Lunsford, Founder and CEO, Tethral
How It Works
TinyFish acts as Tethral's web-side sensory system, adapting to the different ways retailers publish their sale data.
- TinyFish Search resolves a store from nothing but its name and discovers each chain's deals page live rather than relying on a hardcoded URL.
- TinyFish Fetch reads accessible sale pages. For retailers where the circular can be retrieved directly, Fetch renders the page and returns the sale information.
- TinyFish Web Agent handles store-gated circulars. When a retailer requires an interactive store-selection step, Web Agent can select the store by ZIP code, navigate the site, and read the circular.
- Tethral structures the result. The extracted information becomes structured rows that can be matched against the household's meals and used inside the recurring workflow.
Tethral has live-verified full sale listings on retailers including ALDI and Giant Eagle, capturing current price, original price, and percentage off.
For store-gated retailers, Web Agent unlocks another class of circular entirely. In one Food Lion test, a single run selected the store and returned 15 priced items in five steps, in about 90 seconds.
The important part is what happens after retrieval. Tethral doesn't compute or infer the prices. The numbers remain verbatim and traceable to their original source. The model translates the user's intent, while the rest of the workflow runs through a deterministic pipeline with grounded capabilities, authority checks before actions, and receipts afterward. That's what lets a family trust a system acting weekly without supervision.
The team has also proven draft-only cart assembly, filling a guest cart with a user's sale-matched ingredients and always stopping before checkout, and is building per-user store pinning on Browser Context Profiles.
The Results
TinyFish has changed both what the team can retrieve and how much infrastructure it needs to manage to retrieve it.
- A class of data that was previously out of reach: store-gated circulars, where the content only exists after an interactive selection, now return as structured rows.
- One web layer instead of a patchwork of providers: Search, Fetch, and Web Agent cover discovery, extraction, and browser navigation without requiring separate infrastructure for each step.
- Less operational overhead: fewer accounts and integrations mean fewer connections to troubleshoot when something goes wrong.
"It's excellent not just at searching but at navigating what it needed to search. Grocery isn't often sent as raw text, so sales are part of an image. TinyFish was able to make that into a structured dataset without having to attach MCPs, other tools, or add-ons."
- John Lunsford, Founder and CEO, Tethral
What's Next
The team is now working toward a national market baseline for grocery pricing.
Retail chains tend to price by division rather than individual ZIP code, which makes national coverage more tractable than querying every store independently.
The planned system uses Search snippets for change detection, Fetch for retailers with accessible sale pages, and browser automation for store-gated ones.
The goal is a weekly, verbatim, provenance-tagged picture of grocery prices across the US, kept structurally separate from personal household data.
Tethral is also expanding into more consumer services where agents need live discovery before they can coordinate or act.
As those workflows grow, the underlying principle stays the same: the agent handles the user's intent and routine, while TinyFish gives it access to what the web actually says right now.
Building something that needs the web as a reliable data layer?
