AI Actions
The AI Actions column sits on the right side of the dashboard. It holds the tweets you've queued for a response, the drafts the model has generated, plus the controls to post those drafts under your account.
Two action types
Each queued tweet can get one of two actions:
- Reply: a direct in-thread response to the parent tweet.
- Quote-tweet (labelled "Quote RT"): a quote-tweet with your own copy attached.
You pick the action type per row.
Adding a tweet to the queue
Two routes:
- Manual: click the Transfer button on any tweet in the Live Feed or Top Performers. The tweet lands in the queue with the current tone applied.
- Auto: flip on Auto reply generation in the AUTO menu. Patera enqueues and drafts replies for tweets matching your filters as they arrive.
What "auto" actually does
The AUTO menu has four core toggles plus an auto-like section. The four core toggles are independent: none of them requires another to be on. The only thing that disables a row is the plan gate (still loading or a Free plan). Turn on whatever combination fits your workflow.
| Toggle | What it does |
|---|---|
| Auto reply generation | When a tweet enters the queue, immediately calls the model to draft a reply with the current tone. |
| Auto media analysis | When a tweet has photos, runs the multimodal vision pass and folds the result into whichever reply gets generated, by auto reply generation or auto dispatch. Runs standalone: you can auto-analyse media even with reply generation and dispatch off. Spends 1 analysis credit per run. |
| Auto comment analysis | When a tweet enters the queue, reads its replies thread (the same pass as the per-row Analyse comments button below) so the generated reply is informed by the community discussion. Off by default. Runs standalone and draws from the separate comment-analysis quota pool, not the media pool. See Analysing the replies thread. |
| Auto Dispatch | The self-generating master toggle. It drafts any missing reply itself (analysing media when that's on) and sends it the moment it's ready with no review. The dispatched row goes straight into the Execute Queue and rolls through the same worker as Execute All. It needs nothing else enabled first. Off by default: opting in is the signal you're OK with unattended posting. |
Auto Dispatch is the strongest of the four. Read the warning under it before flipping it on: empty or failed generations trigger one regen retry per row and that retry counts against your reply credits. After that one retry, a row that still can't generate (for example, quota exhausted) is dropped rather than re-queued, so a stuck row won't drain your credits in the background.
Auto-like
The fifth section, Auto like posts, is off by default and opt-in. When it's on, executing a reply or quote also likes the target post, posting the like as the X account assigned to that row. A per-type segmented control sets the timing:
| Action type | Options | Default |
|---|---|---|
| Reply | Off / Before / After | Before |
| Quote | Off / Before / After | Off |
Before likes the post first, then sends. After sends first, then likes. The auto-like gets its own row in the Execute Queue and its own toast and it never counts against your reply or analysis quota. Once you enable the master toggle, the choice is remembered.
Per-row controls
Each queued action can be adjusted:
- Switch tone (see Tones): triggers an instant re-generate. The Random tone is a valid choice here. Generate is not disabled and the row shows "Random" as the tone.
- Switch action type between reply and quote-RT.
- Switch the X account that will post it (see Multi-account posting).
- Edit the draft: every generated reply is editable text.
- Regenerate: re-run generation with the same prompt. See Regenerate below.
- Analyse media: manually trigger the vision pass on attached images.
- Analyse comments: read the replies thread and optionally fold it into the reply. See Analysing the replies thread.
- Like the original tweet: a heart on the Original Tweet row likes or unlikes the post this action targets, posting as the selected X account. It runs concurrently (not through the serial worker) and never consumes reply or analysis quota. The filled-red state stays in sync with the same like shown in Live Feed and Top Performers.
The Original Tweet preview inside each row shows the full tweet body by default with no two-line clamp, so you can read what you're replying to. The default is controlled by Settings → AI Settings → Original Tweet preview, which offers Expanded (the default) and Compact (the legacy two-line clamp with a Read more affordance).
Regenerate
Once a draft has text, an explicit Regenerate button appears next to Generate. It re-runs generation with the same prompt and replaces the current draft. It spends 1 reply credit and is gated by the same account, tone, plan and quota checks as Generate. Switching tone also triggers a regenerate, so Regenerate is the way to get a fresh draft when you want to keep the current tone.
Analysing the replies thread
Every row has an Analyse comments button. Click it to read the post's existing replies and distill them into context the model can use. It spends 1 credit from a separate comment-analysis quota pool, not the media-analysis pool (see Plans).
The button is always shown, even when Patera's tracked reply count for the post is 0. The thread is fetched fresh server-side at click time and tracked counts can lag behind X, so the button never hides or blocks on the count alone.
A successful analysis renders a Community context card above the draft, holding:
- A one-line summary of the discussion.
- Topic pills for the recurring themes.
- The top comment (handle, like count, text).
- A sentiment pill.
- An "N replies analysed" count.
After the first run the button relabels to Re-analyse comments, which replaces the existing card with a fresh read.
Folding comments into the reply
Two switches gate whether the analysis (and the raw replies behind it) reach the generated reply. Both must be on:
- The per-row use in reply switch on the Community context card. Running Analyse comments flips this on automatically, since clicking it signals you want the result used. You can turn it back off for a text-only reply on that row.
- The global Context master switch in the column header. This is the on/off for the whole feature across every row.
With both on, the next generation folds in the distilled analysis plus the raw replies. With either off, generation stays text-only.
Empty threads
Analysing a post with zero replies runs no AI, refunds the credit and shows a neutral "No comments found" notice ("This post had 0 comments to analyse, so no AI was run and you were not charged"). It is not treated as an error.
Three ways to send
All three triggers feed the same queue and the same worker. Pick the one that fits the task.
| Trigger | When you'd reach for it |
|---|---|
| Send Reply / Quote RT on a single row | A specific draft you've reviewed and want to send right now. |
| Execute All in the column header | A full queue you've already read through and trust. |
| Auto Dispatch (in the AUTO menu) | You don't want to look at the queue at all. |
Whichever trigger you use, every reply lands in the Execute Queue at the top of the screen. Posts go out serially with a random delay between each (default 0-30s, both bounds editable per tone). Each row posts from the X account assigned to it, so a single sweep can fan out across many accounts in one go.
The serial worker and the random delay are there for one reason: ten posts from the same workspace inside a single second looks like automation and the delay breaks that pattern.
Usage and queue indicators
The column header no longer shows separate replies and media number pills. It now shows a single Usage control. Hover it to open a dropdown with three progress bars, one per quota pool:
- AI replies
- AI-analysed media
- AI-analysed comments
Each bar shows remaining over allowance and is colour-coded by severity (it turns warning, then error, as you run low) with a "Resets ..." footer. On the Free plan the control shows a lock and an upgrade note instead.
Separately, the column scopes by category: the pending / running / completed counts and the replies, analyses, queued filter pills. Click any filter pill to scope the column to that category.
For the global queue across every dispatch path, the Queue pill in the dashboard header is the source of truth. See Execute Queue for the row-by-row anatomy.