Skip to content
GuideForOwnerWith Foxden AdminNeedsan API key from an AI provider

Turn on AI and watch what it costs ​

This page lets you, as an Owner, turn on AI for every app with one provider key in Foxden Admin, set what each app may spend, choose its models, and see where the money goes.

How AI reaches the apps ​

Every app's AI goes through Admin. No app holds a provider key, picks a model or calls a provider itself. Admin holds the keys, decides which model answers each request, counts every call and its cost, and refuses a call that is over an app's limits.

So:

  • With no provider key in Admin, AI is off in every app.
  • An app running alone, without Admin, has no AI.

An app asks for a kind of request, called a tier, and Admin's routing decides which model answers it.

TierWhat it is for
fastQuick, cheap chat completions.
bestThe strongest chat model, for work that justifies the cost.
realtimeLive, in-call assistance where latency matters.
decideA classifier that answers with probabilities, not text.
transcribeSpeech to text, for call and meeting audio.
speakText to speech, for spoken prompts.
embedText to vectors, for hybrid search.

What each app uses them for:

AppUses AI for
ArcticMeeting notes; speaker naming, titles, recaps and Ask; live in-call assist; live call classification (optional)
FennecCall signals, team narrative and deal predictions; the copilot; call transcripts
CorsacFinding the parties on a document and suggesting who to invite
PampasGrammar suggestions; generating forms, response briefs and summaries
CulpeoReplies to website visitors and rewriting search questions; conversation summaries
CapeCall summaries, call transcripts and AI voice prompts
LibraryAsk's answers, follow-up questions and drafts; the brain's briefs and pages; hybrid search (optional)

Add a provider key ​

  1. Open Workspace › AI.
  2. Under AI providers, open Paste a key.
  3. Paste the key under Key.
  4. Leave Provider at Recognise it from the key. For a DeepSeek or Moonshot key, choose the provider: those keys look like OpenAI's.
  5. Select Save key. Admin asks the provider whether it accepts the key before saving it. A refused key is not saved.

The key is stored encrypted and never shown again. The activity log records which provider's key was added, never the key. A key pasted here is used before one set on the server.

You can paste a key for Anthropic (Claude), Groq, xAI (Grok), Google (Gemini), DeepSeek, OpenRouter or Moonshot (Kimi). Groq also serves transcription. Embeddings for Library's hybrid search use an OpenAI or Voyage AI key, and the decide tier a TypeSafe key. Those three are set on the server (OPENAI_API_KEY, VOYAGE_API_KEY, TYPESAFE_API_KEY), not pasted here.

To take a key away, open its row's actions under AI providers and select Remove. From the next request, anything that runs on that provider's models is refused until a key is added again, or the key set on the server is used if there is one.

Check that each app can use AI ​

Under the provider keys, Ready to run? has one row per app on your server:

  • ready: AI is on for the app, it is within its limits, and every provider its models come from has a key.
  • attention: something optional is missing, for example Library's hybrid search without an embeddings key.
  • blocked: AI calls from this app fail now. What to fix names the change; Add a key leads to the key form.
  • not used: the app makes no AI calls.

Select an app's name to open its own AI page.

Set an app's limits ​

Every app has a daily call limit and a spend cap. An app nobody has set limits for runs on the defaults: 500 calls a day and $100 per 30 days. No app is uncapped.

  1. Open Workspace › AI, then Limits and models. Ceilings in force shows every app's limits and how close it is to them.
  2. Select an app to open its page.
  3. Under Ceilings, set:
    • AI for this product: switched off, it refuses every AI request from this app and stops its spend at once. No other app is touched.
    • Calls per day: blank restores the default of 500.
    • Spend cap per 30 days (USD): blank restores the default of $100.
  4. Select Save ceilings.

Choose the model for a tier ​

Each app's page has a Routing section with one row per tier the app uses. A tier nobody has chosen a model for uses the built-in one, so AI works with nothing set here.

  1. On the app's page, find the tier under Routing.
  2. Choose a Model. Only models Admin has a price for are offered, and only from providers that have a key.
  3. Set Max output tokens if you want a lower ceiling on each answer.
  4. Select Save.

To go back to the built-in model, select Reset to built-in on that tier.

Culpeo's fast tier writes the replies to website visitors. Before you route it to a model nobody has marked as evaluated, Admin asks you to tick Route anyway. With such a model, Culpeo withholds every visitor reply and hands the conversation to people. Evaluated models on Limits and models lists the models an Owner has vouched for, per app and tier.

Read what AI costs ​

Workspace › AI shows spend over the last 7, 30 or 90 days (30 by default):

  • Spend, Calls, Average a day and Spends the most at the top.
  • Daily spend: whether the bill is steady or climbing.
  • By product, By model and By provider: where the money goes.

Two warnings appear when they apply:

  • calls have no price: they ran on a model Admin has no price for, so the spend shown is lower than the real bill.
  • calls could not be tied to a person: the spend is counted, but missing from each person's own usage.

Each person sees their own AI calls and cost on their profile. See Edit your profile.

What people see when AI is off ​

When AI is off for an app, its summaries, drafts, suggestions and transcription do not run. That happens when Admin has no provider key, when no model for the tier comes from a provider with a key, or when you switched AI for this product off. Add the key or change the routing, and the app's AI controls come back by themselves once Admin serves them.

An app that is over its daily calls or its 30-day spend cap is refused until the window moves on or you raise the limit.

Troubleshooting ​

"did not accept this key" ​

The provider refused the key. Check that you copied all of it and that it is active in the provider's console, then paste it again. Nothing was saved.

An app shows blocked under Ready to run? ​

Read What to fix on its row. Most often a tier routes to a provider with no key: add that provider's key, or choose a model from a provider that has one.

Spend looks too low ​

Look for the calls have no price warning. Those calls are counted without a cost.

Foxden is open source under Apache-2.0.