A layered free local stack answers cautious long-pause automation; the one cloud sub worth <$30 is Browserbase at $20

Asked (summary)

Find skill for agentic authed browser task execution and low-volume automation.

Asked (follow-up, summary)

Best free, self-host-only stack (no cloud services or accounts) for personal automation with long, very cautious human-in-the-loop sessions — multiple replans and stops. And what is missed by skipping a single cloud tool under $30/month?

Evidence covers 24 rows across official documentation: 11 local browser-automation components, 8 durable-workflow and human-in-the-loop projects, and 5 paid cloud plans under $30/month (the plan rows deduplicate to two distinct offers — Browserbase Developer at $20 and Browser Use Dev at $29). Some feature statements are vendor claims.

The recommended local stack, layer by layer

Hover or tap a layer for its documented evidence and source. Solid arrows: control flow. The model observes and replans only; deterministic Playwright performs known actions.

human gateagent & statebrowser layercontainment & secrets

Playwright MCP records traces with DOM snapshots, screenshots, network activity and console logs — but its allowed/blocked-origin flags are convenience guardrails that do not cover redirects, so the real boundary is the OS firewall. playwright.dev

LangGraph checkpoints failure provenance: if the process crashes mid-interrupt, the graph resumes from its checkpoint with the same context — pauses can wait indefinitely, which no 6-hour cloud session can. docs.langchain.com

Browser Use officially supports Ollama, but its own docs warn small local models can return malformed output and the default context window may be too small — so reserve the local model for observation and replanning. docs.browser-use.com

Temporal offers stronger durability but at single-user scale it is unnecessary operational weight: even its own dev server is "suitable for development and testing only". docs.temporal.io

BrowserSkill (MIT) can reuse the already-logged-in daily browser with built-in human-in-loop takeover — optional, not the safest default, because it widens the blast radius. github.com

Free local vs Browserbase $20 vs Browser Use $29

Concurrency is not the reason to buy cloud; managed runtime, remote reconnectable access, isolation from the workstation, and identity/proxy plumbing are. Cloud never replaces the local checkpointer.

All evidence rows

Component / plan	Set	Documented note	Pages	Source

Method: 24 rows gathered from official documentation of Playwright, Browser Use, LangGraph/LangChain, Temporal, Prefect, Browserbase and related hosts. "Pages" counts documentation pages backing each component's evidence. Plan figures are list prices as published; two synonymous Browserbase rows and three Browser Use Dev rows were deduplicated in the comparison. Long documentation quotes were shortened for space; some capabilities (stealth, CAPTCHA handling) are vendor claims. Nothing here suggests bypassing access controls: at any wall, stop for permitted human action or abort, and respect service terms.

This report was generated automatically by Keenable SELECT at a user's request, from publicly available web sources linked herein. Keenable does not review, verify, or endorse its contents and makes no representation as to accuracy, completeness, or timeliness; AI-based extraction may contain errors. Nothing in this report is investment, legal, financial, or other professional advice. All trademarks and referenced content remain the property of their respective owners; no affiliation or endorsement is implied. To report an error, rights concern, or request removal: legal@keenable.ai.