Summary
PressedMail 1.1.1 is about AI that answers faster and screens that stop getting
in each other's way.
Each AI tool now has a Thinking level. Some providers reason at their maximum
unless told otherwise, which makes every summary, rewrite and phishing check
slow. Pick Low and the same rewrite that took thirteen seconds comes back in
under three. Only the levels your model actually accepts are listed.
The AI Tools page saves from one button in the page header, like every other
settings page. Composer rewrites of long selections stop failing with an empty
response. Cancelling a bulk phishing check, summary or auto-tag run stops it
straight away.
On the rest of the app: it loads once per page instead of twice, a tab left open
through an update recovers by itself, and the speed dial, the list resize handle
and the composer's recipient buttons sit where they should.
Added
- A Thinking level for each AI tool: Model default, Off, Low, Medium and High.
Model default sends nothing, so your provider behaves exactly as before. Lower
levels answer faster. PressedMail asks your model which levels it accepts,
remembers the answer for a week, and lists only those. A level the model
refuses is not saved. Changing the AI provider resets every tool to Model
default. Composer ghost-text suggestions always use the model default.
Changed
- AI Tools saves from one Save button in the page header, with a Cancel beside
it, like every other settings page. It covers the provider, the AI switch,
every tool card and every prompt. The Save buttons on each card are gone, and
so is "Save prompt": prompt edits save with everything else. Leaving the page
with unsaved AI changes now asks first, and a half-edited prompt survives
switching back to the Settings tab.
- The AI provider card no longer has its own Model picker. It edited the same
setting as the Assistant card. Each tool card picks its own model.
- Bulk AI requests give up after 90 seconds. A request that never answered used
to hold the whole run forever.
Fixed
- Composer rewrites of long selections failed with "The AI provider returned an
empty response". A model that reasons by default spent the whole answer
budget thinking and had nothing left to reply with. The budget now grows with
the selection, and when a reply is still cut off the message says to lower
the Thinking level instead of blaming the provider.
- Cancel did not stop a bulk phishing check, summary or auto-tag run. It set a
flag the run only read between emails, so a slow request kept going and a run
whose tab had closed stayed "running" for good. Cancel now stops the request
in flight, a late "done" can no longer undo a cancel, and a run that stops
reporting for ten minutes is marked failed.
- The app loaded twice on every admin page. WordPress asked for the main script
with a version query while every other part of the app asked for it without
one, so the browser treated them as two scripts and ran both. One visible
effect was a composer that closed itself as the page finished loading.
- A tab left open through a plugin update failed to open some screens, and the
browser console showed a 404 for a script such as server.browser-*.js. The
page now reloads once to pick up the new version. An unsaved draft still
asks before anything is thrown away.
- The speed dial and the email list's resize handle were half hidden behind the
composer. The composer sits below both again. The contact and event forms
still sit above the speed dial, so their Save buttons stay clickable.
- Radio buttons and checkboxes in the sweep popup, and anywhere else the plugin
uses the browser's own controls, ignored your color palette and showed the
WordPress admin color instead. They follow your palette now, in light and
dark, including inside dialogs.
- The Cc and Bcc buttons were wider than the Important, Attach files and Media
Library buttons, and Browse contacts had a smaller touch target. All six now
share one size and spacing.
Upgrade Notes
- Nothing to do. Every AI tool starts on Model default, which sends exactly
what 1.1.0 sent.
- If AI answers feel slow, open AI Tools and set Thinking to Low on the tool
that is slow. On a provider that reasons at its maximum by default, that is
the biggest speed-up available.
- A bulk AI run that was stuck as "running" from before the update is marked
failed the first time the Activity panel refreshes.
