For AI agents

Your agent runs the errand.
Fullsy keeps the receipts.

Your agent can request a frozen, verified capture of any URL, on demand. Hand Fullsy a page mid-research and it comes back a fingerprinted Capture Record, filed into a case. "My agent looked it up" turns into something you can check, source by source.

Four ready-made skills · install in a minute

A skill is a set of instructions your agent follows. Read, capture, file, report, with the consent rules built in and every claim tied to a record. The same skills in both Claude Code and Codex. Read them first, they're short on purpose.

research-with-fullsy

Research a question or tear down a set of competitor pages, and capture every source. The answer arrives as a memo where each claim carries a record id.

Claude Code → · Codex →

watch-pages-with-fullsy

Re-capture pages you care about and report what changed. Pricing, terms, claims. Matching fingerprints prove nothing moved. Changes come back with before/after records.

Claude Code → · Codex →

market-with-fullsy

Capture your market and turn it into positioning. The pool to attract, the segment likeliest to buy, and the message that lands. Every competitor claim tied to a record.

Claude Code → · Codex →

sell-with-fullsy

Capture a prospect's site and turn it into a calibrated cold-to-sold plan. Fit, the angle, and who to reach. Every factual claim tied to a record.

Claude Code → · Codex →

Claude Code: save to ~/.claude/skills/<name>/SKILL.md and ask by name (/name). Codex: save to ~/.agents/skills/<name>/SKILL.md and invoke $name. For any other JS-capable agent, paste the skill into the instructions surface it reads.

How it stays yours

Consent first. Audit trail always.

The rules, in order
✓You switch it on. The Agent access switch in Fullsy's Settings, plus one permission prompt. Off is the default, and off means the door doesn't exist.
✓You watch it work. Agent captures open in real, visible tabs, no faster than one every fifteen seconds.
✓Capture, and only capture. It's the same full-page grab you'd trigger by hand, just started by your agent. It can't browse, click, or read your pages. It can only capture the one you point it at.
✓Everything is filed. Each errand becomes a case file, separate from your own capture history, marked agent-initiated.
✓Everything stays local. Records live in your browser. Fullsy has no server to send them to.
Case files

Every errand becomes a case

Agent captures never mix into your own history. They file into a named case, one per errand, on their own shelf. "Download case" hands you the entire errand as a single zip that carries its own proof.

competitor-research-case.zip ├── index.html ← readable case cover, one row per record ├── case.json ← every record, every fingerprint └── 02-competitor-com/ ├── report.html ← readable record: SEO + design + hashes · Pro ├── screenshot.png ← the whole page, stitched ├── document.pdf ← searchable, links click · Pro ├── page.html ← the source as captured ├── page.mht ← sealed one-file archive ├── seo-report.json ← signals + heading tree · Pro ├── design-tokens.json ← their palette, ready to use · Pro └── metadata.json ← the integrity manifest

Fingerprints are SHA-256 hashes recorded at capture time and re-checked every time a record opens. They prove your copy is intact. They are not a legal-authenticity claim.

The bridge

The whole API is four calls

When the toggle is on, every page carries window.fullsy. A small, boring API on purpose. Agents that can run JavaScript in your browser need nothing else installed.

// start a capture of a URL — opens a visible tab, files into a case await fullsy.captureUrl(url, { caseName: 'competitor research' }) // → { ok, capId, status: 'started' } — then poll: await fullsy.getRecord(capId) // → record + fingerprint manifest await fullsy.list(10) // → recent captures, cases marked await fullsy.capture() // → capture the current page

One capture per fifteen seconds. Only http(s) pages. Captures are visible by design. The audit trail is you, watching.

Asked before you ask

Questions

Can a website or an agent capture things without me?

No. Agent access is off until you switch it on, and switching it on is a visible choice with a permission prompt. When it's on, captures happen in tabs you can see, rate-limited, filed into cases you can open and delete. The toggle turns it all off again, and the API disappears from every page.

Which agents work with this?

Any assistant that can run JavaScript in your own browser. Claude in Chrome and tools like it. Agents without browser access can't reach Fullsy at all. The bridge lives in your Chrome, on your toggle, by design.

Is agent capture part of Pro?

Yes. Agent capture is a Fullsy Pro feature, and every install's 30-day trial includes it, so you can run real errands before deciding. After the trial, agent capture and the Records need a Pro license. Your captures and the free tools keep working. On Pro, a downloaded case also carries the report pack for every record. A readable report.html (the SEO and design reports styled like the app, with the fingerprint manifest), a browsable case index.html, plus a searchable PDF, the SEO teardown, and the design tokens.

Where does the evidence live?

In your browser's local storage, like every Fullsy capture. Download a case and the zip is self-contained. The manifest carries every fingerprint, so anyone you hand it to can check the files against the hashes.

Agents move fast.
Fullsy keeps the receipts.

Free capture forever · Pro report packs · 30-day trial included