TagUI: a free RPA tool that runs on your own computer

What TagUI is for, and the maintenance caveat that comes first

TagUI targets repetitive work on a single user's computer. The README lists three surfaces: websites, with native support for Chrome and Edge, desktop applications, and the command line. The intended user is someone who can write a short text file but does not want to build a test framework or an integration platform. The README points beginners at a free course and a YouTube learning series rather than at an SDK.

The caveat sits at the top of the README in a notice: AI Singapore has discontinued maintenance and support of TagUI, and the project continues to be supported by open-source contributors and the community. The repository is not archived, and the last push was on 2026-07-21. That is a community-driven repository, not a vendor product with a support contract. The most recent release listed is v6.110.0 from 2022-06-20, while the README links to a download of v6.114. If you need a vendor to answer a ticket, this is the wrong starting point regardless of how well the flow language reads.

The flow language: identifiers instead of selectors

A TagUI flow is a plain text script. Steps such as click and type act on identifiers, and the README describes four kinds: web identifiers, image snapshots, screen coordinates, and text matched through OCR. That mix is the design decision worth understanding. Most browser automation forces you into DOM selectors. TagUI lets the same step address a button by its visible label, by a cropped screenshot of it, or by a point on screen.

The README gives a login example where the URL is written as a bare line, followed by type email as user@gmail.com and click Log in. The same flow can be rewritten with image files, so type email_box.png as user@gmail.com and click login_button.png work when the page offers nothing stable to anchor on. Data movement is equally terse: table n to forex_rates.csv exports the nth table on a page to a CSV, and a line like top_salesman = [Monthly Report.xlsx]August!E11 reads a cell using ordinary Excel formula syntax. That last choice lowers the learning cost for anyone who already lives in spreadsheets, and it also means the flow language borrows Excel's referencing rules rather than defining its own.

Installing TagUI and running a first flow

The README does not inline install steps. It links to the setup page in the documentation at tagui.readthedocs.io, and the download link for v6.114 points at that same page. Treat the documentation as the source of truth for your platform, because the repository README itself only names Windows, macOS and Linux as supported. The project's own pointer to the download is written as a single link, reproduced here in the form the README presents it:

https://tagui.readthedocs.io/en/latest/setup.html

Once the tool is on the machine, a flow is a text file with a .tag extension. The README's Xero login example is the smallest useful script. The first line navigates to the login page, and the following two fill the fields by their visible labels:

https://login.xero.com/identity/user/login
type email as user@gmail.com
type password as 12345678
click Log in

When a page has no stable label to target, the README shows the same login rewritten against image files instead of text:

type email_box.png as user@gmail.com
type password_box.png as 12345678
click login_button.png

Pulling a table off a page is one line. The README states that n is the nth table on the webpage:

table n to forex_rates.csv

Reading a value out of a workbook uses Excel formula syntax, so a cell reference you already know works unchanged:

top_salesman = [Monthly Report.xlsx]August!E11

Save the file, then run it with the TagUI runner for your platform as described on the setup page. The README does not print the runner invocation itself, so check the documentation for the exact command before you assume a flag exists.

Where TagUI stops being the right tool

The failure mode that matters most is the one the project itself announces. AI Singapore has discontinued maintenance and support. Community contributors continue, and the last push was on 2026-07-21, but there is no commitment to a release schedule, no vendor escalation path, and the newest release named in the repository metadata is v6.110.0 from 2022-06-20. A team that needs a patched dependency on a deadline is exposed.

There are technical limits too. The README names Chrome and Edge as the natively supported browsers, so a Firefox-only estate is out of scope as documented. Image and OCR identifiers trade precision for reach: a screenshot match breaks when a theme, font rendering or window size changes, and the README offers no statement about how it handles that drift. And the security section describes an on-user-computer, on-premises tool that does not store data, which is a genuine advantage for regulated work but also means there is no central bot credential model. The README states that in decentralised bottom-up RPA, bot credentials are not advisable and not needed because users are held accountable. That reasoning holds for a handful of desktop users and stops holding the moment you want an unattended robot running on a server at night.

Alternatives and how their approach differs

The README points at several neighbours, and the differences are structural rather than cosmetic. OpenRPA and OpenFlow are named as leading open-source RPA tools that support enterprise-grade orchestrating and running TagUI robots from their orchestrator. That is the opposite architecture: a central orchestrator dispatches robots, where TagUI's default is a flow file executed on the user's own machine.

Robocorp is named in the same list, again as something that can orchestrate TagUI robots. For Python teams, the README notes that pip install rpa gives access to RPA-Python, a separate package built on the same ideas, so the choice there is between writing flows in TagUI's language and driving automation from Python. There is also a TagUI module for Node-RED for event-driven work with connectors, and a NuGet package for C# .NET. Microsoft Power Automate Desktop users get two-way integration out of the box, which the README frames as business continuity if you switch between the two. Pick on where the control plane lives: TagUI keeps it on the desktop, the orchestrator-based tools move it to a server.

Licence, upgrade cost and what the repository layout tells you

TagUI is Apache-2.0. That permits commercial use and modification, and it includes an explicit patent grant, which matters if you intend to embed the tool in a product rather than use it internally. It is not legal advice, and the licence text in LICENSE.md is what governs.

The upgrade cost is the part to price honestly. Because AI Singapore has discontinued maintenance, upgrades are community work. The README links to v6.114 while the release list in the repository metadata stops at v6.110.0, so anyone tracking versions has to reconcile the documentation download page against the release tags before deciding what to pin. The repository layout is small: .github/, docs/, flows/ and src/, plus a Chinese_Quick_Guide.md at the top level. There is no changelog file at the root, so release notes are the place to look for what changed between versions, and the two most recent entries in that list are the MS Office plugin releases from 2021.

Who should adopt TagUI, and what to check before you do

TagUI fits a specific shape of team: a few people automating their own repetitive desktop and browser work, on machines they control, in an environment where data must not leave the computer. The README's security section is written for exactly that case, and the on-premises default is the strongest argument in the project's favour. The flow language is short enough to learn from the samples in flows/samples, and the MS Office plugins let someone author a robot inside Word with Excel supplying parameters.

It does not fit an organisation that needs a vendor to stand behind the tool. Discontinued maintenance and support is a statement about risk, not about code quality, and the last push on 2026-07-21 shows the community is still active. Before you commit, verify three things: that the setup page at tagui.readthedocs.io lists a version for your operating system, that your browsers are Chrome or Edge, and that the Telegram group is still answering questions about the version you plan to run.

Editorial conclusion

Adopt TagUI if you want an on-premises, file-based RPA tool and you are comfortable that AI Singapore has discontinued maintenance and support, with the community carrying the project forward. Do not adopt it if you need vendor-backed support, a maintained release cadence, or an orchestrator that the vendor stands behind. Before committing, verify that the version you download matches the setup page at tagui.readthedocs.io, check that your target browsers are Chrome or Edge since those are the ones the README names as natively supported, and confirm whether the community is still answering questions in the Telegram group.

Frequently asked questions

Official sources

1. aisingapore/TagUI on GitHub
2. Issues
3. License: Apache-2.0
4. README
5. Releases