yank
An AI assistant that runs your browser — on your own key.
Tell yank what to do and it operates any website for you, using the AI plan you already pay for. And every task it completes is cached as a macro you can replay later for free.
Free forever · Bring your own API key · Runs locally
# Scrape GitHub trending Rust → Notion
navigate("github.com/trending/rust?since=weekly")
wait(selector="article.Box-row")
# for each repo, pull the name + url
for $repo in select_all("article.Box-row"):
$name = extract($repo, "h2 a")
$url = extract($repo, "h2 a", attr="href")
paste_into("notion.so/…", $name, $url)Use the AI you already pay for.
yank is your assistant — not your AI provider. Connect your own key and it drives the browser; you never pay yank for tokens.
Any provider
OpenAI, Claude, Gemini — plus DeepSeek, Qwen, Zhipu and Kimi, or any local / OpenAI-compatible endpoint. Switch in one setting.
No middleman markup
Your requests go straight to your provider on your key. yank takes no cut and adds no per-token fee.
Local-first
Your key, macros, and history live in your browser. Nothing routes through our servers to get a task done.
Run once with AI. Replay forever for free.
Every AI run is captured as a readable macro. When it drifts, AI shows you a screenshot + DOM diff and proposes a patch you review.
AI does it once
Tell yank what you want. It drives the page with browser tools and records the trace as it goes.
yank caches the recipe
The trace becomes a human-readable markdown macro with stable selectors — you can read and edit every step.
Re-run for $0
"Do it again" replays the macro deterministically. No model, no tokens, no per-run cost.
One mental model, three ways to work.
AI-guided run
"Scrape GitHub trending Rust into Notion." AI completes it step by step, recording a trace and producing a macro at the end.
main entry · uses your LLM
Manual record
Hit Record and do it yourself. AI turns your actions into a clean, parameterized macro — great for power users.
for tricky flows
Re-run
Say "again" or hit ▶. Deterministic replay with zero LLM cost — the everyday loop once a macro exists.
daily · 0 LLM
Everything it does — and you stay in control.
Skips re-exploration
Once AI nails a tricky page, it caches the elements that actually worked (keyed by URL). The next visit lands on them directly — no dozens of retries, fewer tokens. Local; selectors only, never your data.
AI-guided execution
Give it a goal; it drives the page with real browser tools — click, type, extract, scroll, navigate — until the task is done.
Readable macros
Every run becomes a markdown macro with stable selectors. Read it, edit it, version it — never a black box.
Zero-LLM replay
Re-run deterministically with no model call. The Nth run costs $0.
Self-heal patches
When a site changes, AI shows a screenshot + DOM/console diff and proposes a patch you accept or reject.
Variables & prompts
AI spots URLs and inputs as variables and turns them into ask steps, so one macro handles many inputs.
Loops over data
Feed a CSV or a page selection; AI infers a "for each row" loop to repeat across every record.
Repetitive browser work that eats your week.
Logged-in scraping
Search LinkedIn / GitHub / Twitter, page through results, paste the top N into Notion or a Sheet — on a schedule.
Cross-platform reposting
Copy a Twitter thread, switch to LinkedIn, format and publish. AI spots the URL as a variable and asks for it each run.
Batch form-filling
Take a CSV or Sheet selection, visit each target page, fill fields, submit. AI infers the "for each row" loop from one recording.
The second run is where everything else keeps paying.
| Claude Skills | browser-use | Bardeen | yank | |
|---|---|---|---|---|
| 2nd run | LLM re-reasons | LLM re-runs | Black-box replay | Deterministic, 0 LLM |
| Nth run cost | N × $0.05 | N × $0.05 | SaaS monthly | $0 |
| Recipe is readable | ✓ markdown | ✗ | ✗ black box | ✓ markdown |
| When it breaks | Re-run LLM | Re-run LLM | Re-record | AI patch you review |
| Local / BYOK | Claude cloud only | ✓ | ✗ | ✓ |
Free does the real work. Pro scales it.
Bring your own LLM key. You only ever pay your provider for the first run.
- ✓ Full core: Modes A / B / C
- ✓ BYOK — any provider
- ✓ Unlimited macros
- ✓ 30 self-heals / month
- ✓ Everything in Free, plus
- ✓ Unlimited self-heal
- ✓ Up to 15 parallel sessions
- ✓ Scheduled runs + session recovery
- ✓ Deep selectors + MCP + import/export
- ✓ Everything in Pro, plus
- ✓ Shared macros + permissions
- ✓ Central key management
- ✓ Audit log
What you get on each plan.
The whole core is free. Pro adds scale, persistence, and connectivity.
| Capability | Free | Pro |
|---|---|---|
| Core — free for everyone | ||
| Browser tools — click, type, extract, scroll, navigate, screenshot, upload… | ✓ | ✓ |
| Page perception — AI snapshots every interactive element before it acts | ✓ | ✓ |
| Shadow DOM + same-origin iframe — sees modern web components | ✓ | ✓ |
| Three modes — AI-guided run · manual record · zero-LLM replay | ✓ | ✓ |
| BYOK + cost tracking — OpenAI / Claude / Gemini / DeepSeek / Qwen / local | ✓ | ✓ |
| Readable macros — markdown, editable, versioned | Unlimited | Unlimited |
| Variables & loops — ask-for-input steps + for-each-row | ✓ | ✓ |
| Password auto-masking — secrets become ask steps, never stored | ✓ | ✓ |
| Context compression — long sessions auto-summarized to fit the window & cut tokens | ✓ | ✓ |
| Page element cache — caches what worked per page to skip re-exploration | 200 pages | 1,000 pages |
| Self-heal — AI proposes a patch when a macro breaks; you review | 30 / month | Unlimited |
| Pro — scale · persistence · connectivity | ||
| Parallel sessions — run multiple automations at once | 1 | Up to 15 |
| Session recovery — resume a run after closing the browser | — | ✓ |
| Scheduled runs — fire macros on a timer (browser open) | — | ✓ |
| MCP — connect & manage MCP servers over HTTP | — | ✓ |
| Macro import / export — back up & share macros as files | — | ✓ |
| Scenarios — named playbooks injected by your trigger | — | ✓ |
| User memory — AI keeps durable facts about you | — | ✓ |
| Bookmarks — read & organize browser bookmarks | — | ✓ |
| Page translation — translate / summarize the current page | — | ✓ |
| Support — community vs priority + early access | Community | Priority |
Questions
How is this different from Claude Skills?
Skills re-run the LLM every time. yank executes once, caches the trace as a macro, then replays it with zero LLM calls — so the Nth run costs nothing.
Do I need an API key?
Yes — yank is bring-your-own-key. You connect any LLM provider and pay them directly, only for the first AI-guided run. Re-runs are free.
Where does my data go?
Macros and history live locally in your browser (IndexedDB). yank talks to your chosen LLM provider and, for Pro accounts, our auth service over HTTPS — nothing else.
What happens when a site changes and a macro breaks?
yank captures a screenshot, the surrounding DOM, and a console diff, then proposes a patch. You review and accept it — no blind re-running.
Stop paying for the same automation twice.
AI does it once. yank caches the recipe. Re-run for free.