Open source · runs on Cloudflare free tier

API-first research agent that returns results in JSON

Claygent-style agent, your stack. Send a prompt + schema — Ferret searches the web, reads pages, and returns data that matches your schema.

ferret · live research loop
💬
ask ferret
You → Ferret
On it — searching the web…
prompt + schema → structured research
developers.cloudflare.com/workers/platform/pricing
Docs / Workers / Platform / Pricing

Workers Pricing

Last reviewed · developers.cloudflare.com

Cloudflare Workers lets you run JavaScript at the edge with no servers to manage. Deploy code close to your users in milliseconds.

The Free plan includes 100,000 requests per day and is ideal for prototypes and side projects. No credit card required to start.

On the Workers Paid plan ($5/month), you get higher limits, longer CPU time, and access to Browser Rendering.

Billing is usage-based beyond the included allotment — you only pay for what you use.

found
G
google.com/search
Workers Paid plan · Cloudflare
$5/month includes 10 million requests. Usage-based billing after that…
Cloudflare Workers pricing explained
How free tier and paid plan request limits work in practice…
ferret · structured answer
🦡

Cloudflare Workers pricing

researched from live web sources

Pricing model Free tier + usage-based paid
Free requests / day 100,000
Paid plan $5 / month

Research that fits how you already work

Search, read, answer. If you can make an HTTP request, you can use Ferret.

Searches, then reads

The LLM picks search queries and pages. No brittle scrapers — the agent navigates like a human researcher.

{ }

Schema-validated JSON

Required fields, types, and enums are checked. One cheap retry if the model drifts off-schema.

Call from anywhere

If it can POST JSON, it can use Ferret — Clay, n8n, Make, coding agents, scripts, or curl.

Highly customizable

Retrofit it with your preferred LLM, SERP, and scraping APIs. Each provider is one function swap via AGENT.md.

Scope to a site

Lock research to a single domain so the agent only searches and reads that site — no wandering off-site.

Budget & deadline

MAX_FETCHES caps tool calls. A 120s deadline makes the agent wrap up instead of hanging your HTTP client.

Tiered page fetch

Native edge fetch first (free), then Browser Rendering, then paid scrapers — only escalate when needed.

Recovers from dead URLs

404s stop the paid cascade immediately. Ferret searches for the live page or another primary source instead of burning credits.

Ship it for free on your Cloudflare account

Clone it, have your agent customize it, drop in your keys, deploy. Free Workers tier is enough to start.

git clone https://github.com/edsmkt/Ferret.git && wrangler deploy
Open the repo