Open-source, self-hosted personal AI agent you run yourself — browser UI, terminal or your chat apps, with memory, MCP tools and scheduled automations.
Rindler
Summary
Rindler signs in to the web portals your team already uses and finishes the task you describe in a sentence — pulling records, downloading invoices, checking statuses, submitting forms.
Screenshots
Description
Most business software assumes an API exists. A great deal of day-to-day operational work happens on websites that have none: supplier portals, court filing systems, insurance payer sites, applicant tracking systems, bank and lender dashboards. Rindler, a Y Combinator-backed company, exists to automate exactly that layer.
One operation
Rindler deliberately exposes a single operation rather than a builder or a flowchart. You describe the task in a sentence, Rindler works out which site the request belongs to, drives a browser on its own infrastructure, and hands back what the site said. There is nothing to open or close and no supported-sites list to consult first — any website your team already has access to can take a request.
How it avoids the usual fragility
The common approach to web automation does the discovery work live on every run: fetch the page, read the DOM or accessibility tree, guess selectors, click, re-read and recover. That repeats on each execution and breaks whenever a layout shifts. Rindler instead learns a site once and maintains that mapping centrally, so when a portal changes its interface the fix is deployed for everyone rather than debugged by each customer. The company publishes benchmarks claiming roughly 4x faster task completion and 3x lower cost than open-source browser-agent alternatives.
Control and visibility
Logins are stored encrypted and scoped to the individual site they belong to. Every step and screen of a run is captured so an operator can see what actually happened. Permissions are set per action and per site with three settings — allow, ask first, or never — so consequential steps such as submitting a form or making a payment can be gated. Rindler also documents an MCP access channel, with docs covering MCP setup and browser MCP for programmatic use from agents.
Pricing
Billing is per finished task rather than per step or per minute: one run equals one completed task equals one dollar. Starter is $100 a month for 100 runs against a single login-gated site, set up with the Rindler team, and begins with a 7-day free trial with nothing charged until day 8. Teams is $1,000 a month for 1,000 runs across ten login-gated sites and adds saved tasks on a schedule plus a shared team workspace. Enterprise is custom-priced for portal work spread across many sites, with extra capacity, dedicated onboarding and an annual agreement billed monthly.
Who it is for
Rindler targets procurement and accounts payable, government and courts, healthcare and insurance, HR and recruiting, banking and lending, and commerce, travel and property teams — anywhere a person currently logs into a portal by hand and repeats the same handful of actions.
Reviews
Similar App Suggestions
A pay-per-call tool layer for AI agents: one balance and one credential across 1,700+ tools from 55+ providers, reachable as a skill, an MCP server or a CLI.
Watches how your team actually works, learns a back-office process from one screenshare, and ships a governed agent that replays every decision it makes.
Mac app that exposes your browser, CLI, files and saved logins to Claude, Codex or any MCP agent — with an approval gate on every request.
Describe an internal tool in plain English and Prized builds and deploys it, scoped to connectors an admin approved and logged end to end.