Large workloads and scaling

Running Aimogen Pro at volume - bulk generation, archive-wide editing, large knowledge bases and busy chatbots.

Bulk generation at volume#

Understand the request count first. A rule generating a title, an outline, six sections, an introduction, a conclusion, an excerpt, categories, tags and two SEO fields makes around fourteen requests per post — before images.

At one post per run, every 15 minutes, that is roughly 1,300 requests a day.

Practical measures:

Measure	Effect
Start at one post per run	You learn the real cost per post before scaling
Use a mid-tier model for the body	The single largest cost lever
Use the cheapest capable model for titles, excerpts and meta	They are a few dozen tokens each
Raise `Delay Between API & Scraping Requests (ms)`	Avoids the 429s that cause retries
Add more API keys	Spreads the rate limit. See key rotation
Enable Reliability Mode	A provider failure does not stop the run
Run through WP-CLI	No web timeout
Schedule for off-peak hours	Less contention with visitor traffic

Editing an archive#

The most expensive single operation available. A thousand posts, several elements each, is a large bill and a long run.

Take a backup. There is no undo.
Test on three posts with Publish as New Post After Editing.
Measure the cost in System & Logs › Usage Logs.
Multiply by your post count. Decide whether the model choice survives that arithmetic.
Batch. 10–25 posts per admin run, or use WP-CLI for larger batches.
Filter narrowly. Custom field or state filters keep you in control of what is touched.

bash

wp post list --category=5 --format=ids --posts_per_page=-1 \
  | tr ' ' ',' | xargs -I {} wp aimogen-edit {} 7

See Editing existing content in bulk.

Large knowledge bases#

Indexing is one API call plus a store write per document. A thousand posts is a thousand of each — cheap in money, slow in time. Index in batches.

Turn off Optimize The %%post_content%% Shortcode Using AI for a large initial index; it adds a full AI rewrite per document.

Retrieval cost scales with usage, not index size. A million-document index costs the same per query as a thousand-document one. What costs is the number of chunks injected — every chunk is input tokens on every affected request.

Vector store choice matters at scale. All four options handle a typical WordPress site comfortably. If you are indexing hundreds of thousands of documents, review pricing and query performance for your chosen store.

See Knowledge base and RAG.

Busy chatbots#

Latency dominates. Choose a fast model.

Context is resent every turn. With 20 messages of history, message 21 sends all 20 again. Reduce Max Chat Messages To Send As API Context to something realistic — ten is usually plenty.

Retrieval multiplies that. Five chunks of 500 tokens is 2,500 extra input tokens per message.

Exclude admin-ajax.php from page cache, or visitors receive each other responses.

Rate limit. Ten messages per minute is invisible to a real visitor and stops abuse.

Set guest credits. Before it is public.

Many agents#

Agents are the most resource-intensive feature: every step is a request, and steps are dispatched through cron.

Keep Max Steps low
Do not run many agents simultaneously — they compete for the same cron cycles
Do not enable delegation unless you have designed for it
Watch aiomatic_agent_runs growth on a site running agents continuously

Database growth#

Table	Grows with	Control
`aiomatic_logs`	Every request	Log retention
`aiomatic_logmeta`	Detailed tracking. Fastest	Turn detailed tracking off in production
`aiomatic_agent_runs`	Agent steps	Periodic cleanup
`aiomatict_shortcode_rez`	`[aicontent]` usage	Bounded by cache duration
`wp_posts`	Generated content and plugin post types	Expected

A large wp_options table slows every request, and the plugin stores several large options. Keep an eye on it on a site with many rules.

Hosting#

At volume the constraints are outbound HTTP concurrency, PHP process limits and execution time. Shared hosting will struggle with sustained bulk generation.

Signs you have outgrown your hosting: runs consistently dying mid-way, stale run locks appearing regularly, and 429s from your host rather than the provider.

Measuring before scaling#

The honest sequence:

Run the smallest realistic version.
Read System & Logs › Usage Logs for actual cost and token counts.
Read Activity Logs for failure patterns.
Multiply by your intended volume.
Decide whether the numbers work before turning it up.

Most cost surprises come from skipping steps 2 to 4.

Still stuck? Open a support ticket and include the diagnostics from Aimogen Pro › System & Logs › System Info.