Skip to main content

AI Providers

Several SyteOps features call AI models — LinkCentral keyword enrichment and cross-link scoring, Content Pipelines drafting, GEO readiness analysis, content ingest mapping, social content generation, and more. SyteOps supports six providers and lets each feature pick its own provider, model, and token budget so you can match the model to the job.

Supported providers​

ProviderTypical useNotable features
OpenAIGeneral chat and reasoning models (GPT-4o, GPT-5, o-series)Mature API; broad model catalog
AnthropicClaude family (Opus, Sonnet, Haiku)Strong instruction-following; long context
OpenRouter (recommended)Aggregated access to models from most major providersOne key unlocks every listed model — including Perplexity Sonar
GeminiGoogle's Gemini 1.5 / 2.x / 3.x familiesLong context; native Google integration
PerplexitySonar models with built-in web searchReal-time web context for destination-site lookups

Why OpenRouter is the recommended default: one API key gives you access to models from every other provider. If you're not sure which provider to use for a given feature, start with OpenRouter and pick a model from the provider you prefer.

Perplexity's web-search exclusivity: the Context AI step in LinkCentral keyword enrichment requires a model that can perform web search. Only Perplexity and OpenRouter can serve those models. SyteOps automatically filters the Context AI provider dropdown to show only providers that can access them.

Where API keys live​

  1. Open the System / API tab in SyteOps settings
  2. Find the API Keys tile
  3. Paste each provider's key into the matching field
  4. Click Save Changes

All five keys are encrypted at rest. They are never displayed in plaintext after save — you see a masked value.

Test a saved key: each tile has a Test key button (shown once a key is saved). It makes a quick live call to the provider and reports back — "Key valid — N models available", or the provider's error message if the key is rejected. Useful for providers that don't expose a balance you can sanity-check against.

SyteHero image AI (fal)​

Below the provider tiles, an Image AI group holds one more tile: SyteHero image AI (fal). SyteHero generates and modifies pictures through fal, and normally holds that key itself. Turn Manage this key here on and SyteOps holds it instead — you paste the fal key once, on this card, and SyteHero borrows it.

This tile does not need FlowMattic. On a site without FlowMattic the System / API tab still shows it, on its own and with its own quick-navigation pill named after the tile, so you can set, change or clear the key and switch management off; the rest of the AI API keys stay hidden until FlowMattic is installed.

Update SyteHero to version 1.0.139 or newer before you turn the toggle on. An older SyteHero does not know to ask, so the toggle would simply do nothing while its own key controls stayed live. Nothing breaks either way — SyteHero falls back to whatever key it already has — but the setting will not take effect until it is updated.

What changes while the toggle is on:

  • SyteHero uses the key from this tile for every image call, and never stores a copy of it.
  • On SyteHero's own Integrations page, the fal card's key field, enable switch, clear checkbox and Test Connection button are disabled, with a "Managed by SyteOps" note and a link back here. Its own stored key is left untouched and comes back the moment you turn the toggle off.
  • Each capability row on the tile — generate a picture, remix a picture, picture to video, remove background, upscale, extend edges — pins a model id for that job. When SyteHero is active, each row is a dropdown of the models SyteHero supports, showing each model's name and cost ($ to $$$$). Choose Let SyteHero choose to leave that job to SyteHero. Until you save your own picks, value-for-money models are preselected: good quality at a modest per-image price. Save to use them, or change any row. Without SyteHero active, the rows are plain text fields for a model id.

The key is encrypted at rest, never shown again after save ("Leave blank to keep current key" plus a Clear key on save checkbox), and excluded from configuration exports and backups, exactly like the provider keys above.

Reading the balance. The same fal key you paste above can generate images fine while still being refused when SyteOps tries to check your account balance — fal only allows balance reads from an Admin key on the account that pays the bill, not from a Team key. If the balance explains that it can't read your account, put an Admin key from the client's fal team (fal.ai/dashboard/keys) in the fal key field above — fal has no way to scope a key any narrower than that, so the same key that generates images also reads the balance once it's an Admin key.

fal billing key (deprecated). An older, separate key some sites used to work around this before an Admin key was known to cover both jobs. It no longer unblocks anything a fal key above doesn't already fix, so it appears on this tile only for a site that already has one stored — clear it there once the fal key above is an Admin key. It is encrypted, masked, and excluded from exports the same way every other key here is.

Default collection. SyteOps Admins see a read-only Default collection line under this tile once one is set in SyteHero's AI Studio — the collection new pictures are filed into by default. It shows the collection's name, with a link to open SyteHero's AI Studio Assets view. Nothing here can create or change a collection; that stays entirely inside SyteHero's own AI Studio.

Other service keys​

Some services SyteOps talks to are not AI providers, so their keys live further down the same System / API tab, with the other service credentials rather than in the API Keys tile:

  • Erlin API Key — for Erlin AI, an AI-search visibility service. Generate the key in Erlin under Settings → Accounts → MCP; it begins with ek_live_. The field simply stores the key — nothing in SyteOps contacts Erlin yet.

Like the provider keys, it is encrypted at rest, masked after save, and excluded from configuration exports.

Per-provider default model and token limit​

Each provider tile on the API Keys card has its own Preferred model and Max tokens fields, right below the key field:

  1. Enter and save the provider's API key.
  2. Click Fetch Models to load that provider's live model list, then pick a model (or choose Custom to type a model ID by hand).
  3. Set Max tokens, or click the refresh (↻) button to snap it to the selected model's maximum.

These give you a single, provider-level default — handy when you want every automation that uses a provider to share the same model and token ceiling. Each field also has a copy icon that copies its FlowMattic variable (wrapped in {{ }}) so you can drop it straight into a workflow:

  • {{syteops_std_systm_openai_model_001_preferred}} / {{syteops_std_systm_openai_model_001_max_tokens}}
  • …and the matching anthropic, openrouter, gemini, and perplexity variables.

The model and token-limit fields appear on a provider tile once that provider's key is saved (so the model list has a key to load with).

Used as a fallback: when a feature that runs AI in SyteOps (e.g. LinkCentral cross-linking and keyword matching) has no model of its own set, it falls back to the matching provider's default model and token limit here — so you can set a sensible default once per provider instead of per feature.

Site default provider​

The API Keys card carries one Site default provider select, above the provider tiles. It lists only providers whose key is saved, and it decides what happens for any AI area you have not pointed at a provider of its own: that area uses the site default, along with that provider's Preferred model and Max tokens from its tile.

Leave it on Auto-pick and SyteOps chooses for you — OpenRouter when its key is saved, otherwise the first provider in the list that has one. The hint under the select names the provider auto-pick would land on, so the fallback is never a mystery.

Two things to know:

  • Setting it changes nothing you have already configured. An area with its own provider always wins; the site default only fills in for areas you left blank.
  • It fails safe. If the chosen provider's key is later cleared, the site default is ignored and areas fall back to auto-pick rather than stopping — the select shows the stale choice flagged "no API key" so you can change it.

Where an area is running on the site default, SyteOps says so wherever a provider·model pair is shown — the AI Models card's own line (below), and the provider line on the Review Portal's Remix and Agent prompts panels — as "Anthropic (site default)" with the model that will actually run.

Per-feature provider and model selection: the AI Models card​

The Content Pipelines → AI Models view carries one AI Models card, one row per feature, named by what it powers rather than by an internal area slug:

  • Article writing & remix (Content) — AI-suggested categories, tags, and meta descriptions for content sources and the Review Portal, plus the Text AI Remixer. See Content Pipelines.
  • Answer-engine analysis (GEO) — AI readiness analysis that scores how well a post is positioned for AI answer engines. See Content Pipelines.
  • Ingest field mapping (Ingest) — proposes how an inbound content-source payload maps to your post fields when you set up a new content source. See Content Pipelines.
  • Social posts (Social) — AI-generated social posts from your published content. Its provider/model/max-tokens fields render on this card and save with it; the Social Publishing view keeps its own voice profiles, destinations, and enable toggle.
  • Image alt & captions (Image SEO, carries a Vision badge) — Generate SEO with AI in the Review Portal. It writes an image's alt text, title, and caption from the picture itself rather than the surrounding words. It uses that provider's key on System / API, not SyteHero (SyteHero is Modify with AI). This area must point at a model that accepts image input; a text-only model is refused rather than asked to guess, so a provider that cannot take an image shows disabled here. See Review & Publish Your Post.
  • Agent prompts — turns a page reviewer's notes into an implementation prompt. A Default approval mode setting (Propose only, Approve each change or Approved batch; Approve each change by default) chooses which mode the portal card starts on. A per-site preamble template (with a Restore default option) sits beside it, in its own card just below.

Two more rows are status-only, with no controls of their own:

  • Lead scoring — shown when the Leads module is active. Links straight to the Leads page, where its own AI provider/model fields live (used for Auto-map with AI).
  • LinkCentral — shown when the LinkCentral integration is on, as three rows (Cross-Link AI, Keyword AI, Context AI). Each links to the LinkCentral integration page's Enrichment Settings on System / API, where its own controls live.

Reading and using a row​

Each row shows:

  • A readiness pill, with a tooltip naming the provider it is about — Ready (a provider is chosen, explicitly or inherited, and that provider has an API key on System / API), Needs key (a provider is chosen but that provider's key is missing — or, for Image SEO, the resolved model lost image capability), or Not configured (nothing resolves at all — only happens when no provider anywhere on the site holds a key). This applies to the LinkCentral rows too, reading their own provider choice the same way every other row does.
  • A "Powers: …" line — one sentence naming what the area actually does.
  • "Uses site default (Provider · Model)" — shown whenever the row has no explicit provider of its own, with the resolved pair that will actually run. An Override toggle sits beside it.

Turn Override on to reveal the provider and model selects and pick your own. Turn it back off and save: the row's provider and model are cleared back to '', which is exactly what "inherit the site default" means everywhere else in SyteOps — the change takes effect on the next Save, not the moment you flip the toggle.

An Advanced disclosure under each row holds Max tokens — a per-area setting, editable independent of Override. To use a model that isn't in the fetched list, choose Custom in the model select (still inside the main row, not inside Advanced) and type the exact model ID.

The card's header links to API keys & site default, on System / API, for the keys and the site-default provider setting themselves. On a site without FlowMattic that section is not shown, so the card instead points to the Systems / API tab, where FlowMattic can be installed.

Image generation & modify (SyteHero)​

At the foot of the same card sits a read-only Image generation & modify (SyteHero) row. It reports whether SyteHero is installed and active, whether its image-AI key is configured, and — once you've checked it from the balance popup or the Review Portal's own image tools — its cached credit balance. When available, the row also names this month's image-generation spend — something like "This month: $12.50 · 3 generations (fal list price)" — SyteHero's own local record, not a live vendor read, so it appears with no extra click. If any of that spend is an estimate rather than an exact figure, the row says "fal prices & estimates" instead of "fal list price". Once SyteHero confirms a picture's ACTUAL billed charge with fal (rather than its own list-price estimate), a fully-confirmed month instead says "fal billed", and a month with some confirmed and some not says "fal billed + list prices"; if a generation went through with no price on record at all, the row says so ("N not priced") rather than understating the total. When SyteHero can also read your fal TEAM's total usage, the row appends "fal team total this month: $X" — the whole team's spend across every site sharing that key, alongside this site's own figure. When SyteHero is active, the row links straight to Open SyteHero settings; when it isn't, the row simply says image AI needs SyteHero. This row has no controls of its own — it powers picture remix, generate-new-picture, and the listing-image watermark repair. Where its key and models are configured depends on one setting: SyteHero's own settings page by default, or the SyteHero image AI (fal) tile on System / API when you have turned management on there, in which case the row says so and links straight at the tile. If a balance has never been read successfully because your fal key can't check it, the row shows the same Admin-key hint the balance popup shows, instead of staying blank.

The other way to set AI configuration​

Some integrations and modules also offer a Configure AI Provider modal — click it on the integration card (Integrations tab) or the module dropdown (Modules tab). It opens with provider, model, and max-tokens fields and saves directly via AJAX, without navigating to the feature's own settings tab. It is a separate path from the AI Models card above — a feature whose controls already render on that card (Content, GEO, Ingest, Image SEO, Agent prompts, Social) is set there instead.

Live balance display​

Each provider's card on the API Keys tile shows your current balance (or credit/usage, depending on what the provider returns):

  • The balance row only appears once you've saved a key for that provider — providers with no key entered don't show a balance (and won't surface a balance-lookup error).
  • Click the refresh icon to re-fetch the balance.
  • If a provider doesn't expose a balance endpoint, the field shows "—".

Use this to catch low-balance states before an enrichment run or content-pipeline batch exhausts your credit.

SyteHero's row also shows this month's spend. Below its balance figure, a line reads "This month: $X · N generations (fal list price)" — the total spent generating and remixing pictures on this site so far this month, from SyteHero's own local record. Click Previous months to expand up to five more months. If some of a month's spend is an estimate rather than an exact figure, the line says "fal prices & estimates" instead; once fal has confirmed the actual charge for every priced generation that month it says "fal billed" instead, or "fal billed + list prices" when only some are confirmed; a month where nothing was priced yet reads "price unknown" rather than a misleading $0.00. This line appears wherever the balance figure does — the branding modal and the Image generation & modify (SyteHero) row above.

Provider branding modal​

Clicking the plugin logo in the top corner of the settings screen opens a small information panel with:

  • Link to the provider's primary documentation
  • Link to the provider's pricing / models page
  • The SyteOps-side field the provider's key maps to (useful when you're verifying keys by provider)

Model max-tokens behavior​

  • When you select a model, SyteOps auto-populates the Max Tokens field with the model's maximum output size (when the provider exposes per-model metadata).
  • A hint below the field reads "Model maximum: X tokens" so you can see the ceiling without opening provider docs.
  • The refresh (↻) button next to the Max Tokens field resets the value to the model's maximum.
  • If you use a custom model ID (a model not in the dropdown), the max-tokens ceiling falls back to the provider's default.

Model recommendations by use-case​

Use-caseModel classExamples
Keyword generation (LinkCentral Keyword AI)Fast, lightweight chatGPT-5.4 Mini, GPT-4o Mini, Claude Haiku
Cross-link relevance scoringStandard chatSame as keyword, or a step up like GPT-5.4
Web-search context (LinkCentral Context AI)Search-enabled (Sonar family)Sonar, Sonar Pro
Drafting / longer-form contentCapable chat or reasoningClaude Sonnet, GPT-5, Gemini 2.5 Pro
Structured extraction / classificationFast, lightweight chatHaiku, Mini variants

Reasoning / thinking modes add latency. For short tasks like keyword generation, a plain chat model is typically better.

Troubleshooting​

"Models not loading" in the model dropdown

  • Confirm the provider's API key is saved (the field should show a masked value, not be empty).
  • Click Fetch Models on the AI area to force a refresh of the cache.
  • Provider model caches are cleared automatically on plugin update; other providers refresh on demand.

"Balance shows —"

  • The provider either doesn't expose a balance endpoint, or the response format changed. Try the per-provider refresh icon.
  • Check the provider dashboard directly as a fallback.

fal balance shows "This fal key can't read the balance…"

  • Your fal key can generate images but is a Team key, and fal only allows balance reads from an Admin key on the account that pays the bill. Put an Admin key from the client's fal team in the fal key field on the SyteHero image AI (fal) tile — see SyteHero image AI (fal) above; fal has no way to scope a key any narrower than that, so nothing else about image generation changes. If this site already has a fal billing key stored from before, clear it there too — it's deprecated and no longer needed.

"Perplexity doesn't appear as a Context AI provider"

  • The Context AI dropdown is filtered to Perplexity-capable providers (Perplexity, OpenRouter). If you selected a different provider for Keyword AI, the Context AI dropdown still shows only the eligible providers — this is by design.