Versions
Which Open WebUI and OpenClaw versions the integration is tested on, the minimums, and what an older release loses, with the CI runs that show it.
Applies to: Pipe, Action, Filter
Tested and minimum
| Minimum | Tested in CI | |
|---|---|---|
| Open WebUI | 0.12.0 | 0.12.5 |
| OpenClaw Gateway | 2026.9.3 | 2026.10.5 |
| Python inside Open WebUI (what Open WebUI ships) | 3.11 | 3.11 and 3.12 |
Python for install.py | 3.10 | 3.10 to 3.13 |
The "Tested in CI" versions are the ones the end-to-end suite runs on for every pull request (e2e/versions.env). The minimums come from the same suite run once on each older release; the results are below.
Below the minimums the chat itself still works, but some features break (the table below). The worst cases: on Open WebUI 0.11.0 and older, tool cards keep spinning; on Open WebUI 0.11.1, and on OpenClaw 2026.9.1 or older, cron replies never appear.
What each feature needs
| Feature | Needs | On an older release |
|---|---|---|
| Tool cards (live, native) | Open WebUI 0.11.1 | 0.10.2 and 0.11.0: a card keeps spinning after its tool has finished, because 0.11.1 changed how Open WebUI pairs a call with its result and the Pipe follows 0.11.1 |
| Replies the agent starts (cron jobs, sub-agent results, restart recovery) | Open WebUI 0.12.0 | 0.11.1 and older: the reply is never written to the chat. The writer needs the chat row lock 0.12 added |
| Status dialog styling | Open WebUI 0.12.0 | Older releases lack a few of its CSS classes: slightly different spacing and text size, nothing missing |
| Status button in the composer on chat open | Open WebUI 0.12.0 | Appears after the next reply instead |
| A reply that finishes while a tab is loading the chat | Open WebUI 0.12.4 | 0.12.0 to 0.12.3: that tab can show the reply cut until it is reloaded. Those releases replace what reached the page during a load with what the load fetched; 0.12.4 keeps it |
| Cron replies, and finishing a reply a Gateway restart cut off | OpenClaw 2026.9.3 | 2026.9.1 and older: a cron run in the chat's session is not recognised, so its reply is never shown; and in the restart test, a reply cut by a Gateway restart did not finish |
| Rate limits for a Gateway that only has a Claude CLI login | The claude-cli-usage Gateway plugin | OpenClaw 2026.9.1 and later report no Anthropic windows for that login; the dialog says so (Status button) |
Test runs
Each row is the full end-to-end suite on one pair, on GitHub's runners. The suite was not changed to suit older releases: a failure is the integration's behaviour on that release.
Open WebUI, with OpenClaw 2026.9.3
| Open WebUI | Result | Failed tests | Run |
|---|---|---|---|
| 0.10.2 | 11 of 16 pass (1 skipped) | tool card stuck spinning; agent-started replies; sub-agent drawer (its parent reply never finished); Status CSS classes | run |
| 0.11.0 | 11 of 16 pass (1 skipped) | the same as 0.10.2 | run |
| 0.11.1 | 13 of 16 pass (1 skipped) | agent-started replies (two tests); Status CSS classes | run |
| 0.12.0 | pass | run | |
| 0.12.2 | pass | run | |
| 0.12.3 | pass | run | |
| 0.12.5 | pass | run |
OpenClaw, with Open WebUI 0.12.0
| OpenClaw | Result | Failed tests | Run |
|---|---|---|---|
| 2026.8.35 | 14 of 17 pass | cron replies (two tests); a guest's reply cut by a Gateway restart did not finish | run |
| 2026.9.1 | 14 of 17 pass | the same as 2026.8.35 | run |
| 2026.9.3 | pass | run | |
| 2026.10.3 | pass | run | |
| 2026.10.5 | pass | every pull request |
Keeping this page true
- A unit test fails when the "Tested in CI" row here or in the README differs from
e2e/versions.env. - To re-measure, run the end-to-end workflow by hand (Actions > E2E > Run workflow) with the
owui_imageoropenclaw_versioninput set to the release, and update the tables. Do it when an upgrade touches one of the features above.