Tutorial 09

Connect a cloud AI model

Why this matters

Agents read context files. Tools let them act on the world. But the part that actually thinks — that strings together a sentence, weighs the options, drafts the email — is the AI model.

Auxot doesn’t ship with one. It connects to whatever AI model you bring: an OpenAI account, an Anthropic account, your own GPU running locally, a CLI tool already on your laptop. You decide which provider does your thinking, and Auxot routes work to it.

Cloud providers are the easiest place to start. Sign up with OpenAI or Anthropic, get an API key, paste it into Auxot, done. Your agents start working within thirty seconds.

(If you’d rather run AI on your own hardware so no data leaves your building, that’s Tutorial 10. The two paths aren’t mutually exclusive — many teams connect both, with cloud as the fast default and local GPU as the privacy-protecting fallback.)

Today, connect one cloud provider. Tomorrow, your agents have a brain.

You connect a cloud AI provider — OpenAI, Anthropic, or both — by giving Auxot the API key from your account with that provider. From then on, Auxot routes your agents’ work to that provider. You stay in control of which models show up in the picker and which models each agent prefers.

Quick start

1. Get an API key from your provider. OpenAI: log into platform.openai.com → API Keys → Create new secret key. Anthropic: log into console.anthropic.com → API Keys → Create Key. Copy the key value (you’ll only see it once).
2. Sign in to Auxot in your browser.
3. Open Providers — click Settings in the left menu, then Providers.
4. Add the provider — click + Add Provider, pick OpenAI or Anthropic, give it a clear name (something like “OpenAI production”), paste your API key, and create.
5. Pick which models to surface — Auxot fetches the available models from your provider. On the provider’s detail page, check the boxes for the models you want available in the agent picker (start with the latest one).

Done? Open chat with any agent, pick the new model from the model selector, send a message. You’re now running on cloud AI.

The agent can do that?

You don’t have to manage providers entirely on your own. The Admin Agent can recommend models, configure fallback chains, and diagnose when something’s wrong. Three prompts that put its provider knowledge to work.

1. Have the Admin Agent recommend the right model per agent

Open chat with the Admin Agent and ask:

Look at all the custom agents in my account. For each one, recommend which AI model from my connected providers it should use as its preferred model — and tell me why. Match the model to the agent's job: a customer-support drafting agent doesn't need the most expensive model; a contract-review agent probably does.

Why it’s non-obvious: Most accounts use the same default model for every agent because nobody wants to think about it. The Admin Agent can read each agent’s job description, weigh it against the models you’ve surfaced (cost vs. capability vs. context window), and recommend a per-agent default that saves money on simple work and uses the better model where it matters. Five-minute task; five-thousand-dollar/year decision in a busy account.

2. Configure fallback routing

Same chat, follow-up:

I want my agents to keep working if any one provider has a bad day. Look at what I've connected, recommend which provider should fall back to which (and under what conditions — offline, over quota, queue too deep), and walk me through setting it up.

Why it’s non-obvious: Auxot’s queue-don’t-fail design means agents wait gracefully when a provider is having issues. Adding fallback routing turns that quiet wait into invisible recovery — the agent’s work runs on the backup provider before the user notices. Most people skip configuring this until something breaks. The Admin Agent can recommend the right rules in one conversation.

3. Diagnose a flaky provider connection

When something feels off:

My [provider name] keeps showing as having issues — slow responses, occasional errors, or it just feels off. Walk me through what to check, in order, and tell me which is most likely the cause based on what you can see in System Health.

Why it’s non-obvious: “Why is my AI slow?” is one of those questions where the cause could be in three different places — your provider’s status (their problem), your account’s quota (your problem), or Auxot’s connection settings (Auxot’s problem). The Admin Agent has access to System Health, the queue depth, recent errors, and your provider configuration. It can triage the question in fifteen seconds with a single prioritized list. Better than hunting across three browser tabs trying to figure out where to look first.

Go deeper

Walkthrough

Step 1: Get an API key from your provider

If you’ve never used OpenAI or Anthropic’s API directly, this is the part that takes the longest. It’s still under five minutes.

a. OpenAI

1. Go to platform.openai.com and log in (or create an account).
2. Add a payment method under Billing → Payment methods. OpenAI doesn’t allow API key creation without one. Most teams pre-load $10–$25 to start.
3. Click API Keys in the left menu (or visit platform.openai.com/api-keys).
4. Click Create new secret key. Give it a name like “Auxot — production” so you’ll know what it’s for later. Copy the key value (it starts with sk-...). You won’t see it again — paste it somewhere safe immediately.

b. Anthropic

1. Go to console.anthropic.com and log in (or create an account).
2. Add a payment method under Plans & Billing. Anthropic also charges per usage — typical starting balance is $5–$25.
3. Click API Keys in the left menu.
4. Click Create Key. Name it “Auxot — production” (or whatever helps you recognize it later). Copy the key value. Same rule: you won’t see it again.

Tip: Both providers charge per call, not per month. Costs scale with how much your agents do. A small team running a handful of agents typically spends $5–$50/month. Heavy automation can run higher. Both providers let you set spending limits in their billing settings — do this so a runaway loop doesn’t surprise you.

Step 2: Sign in to Auxot

Open Auxot in your browser and log in. You’ll need to be an org admin for the next steps; non-admin users can see providers but can’t add or edit them.

Step 3: Open Providers

Click Settings in the left menu, then Providers. The page lists every AI provider currently connected to your account.

The page’s empty-state copy says it well: “No providers yet. Add one to connect your agents to an AI backend like Claude, GPT, or your own hardware.”

If you already see providers in the list (someone connected one during account setup), you can still add another. Auxot supports multiple providers running side by side — useful for fallback routing, which Step 5 covers.

Step 4: Add the provider

Click + Add Provider. A modal opens with the title “Add Provider” and the description “Choose a provider type and give it a name. You’ll configure the details on the next screen.”

Three things to fill in:

• Provider type. Pick OpenAI or Anthropic. (You’ll see two other options — GPU and CLI — for local hardware setups; those are Tutorial 10 territory.)
• Name. What you’ll recognize this provider by. Something like “OpenAI production” or “Anthropic — Claude account.” Naming matters when you have several connected; “My OpenAI” is fine for one, but if you connect two OpenAI accounts later (production and a separate dev one, say), descriptive names save confusion.
• API key. Paste the key you copied in Step 1. Auxot encrypts it before storing — even Auxot itself only ever sees a hint (first/last few characters) after this point.

Click Create.

Step 5: Pick which models to surface

Auxot drops you on the new provider’s detail page. Behind the scenes, it’s already calling OpenAI or Anthropic to ask “what models does this account have access to?” The list populates within a few seconds.

You’ll see a section titled something like “Available models” with the helper text “Refresh the list from your provider, then choose which models appear in the chat selector.”

Two things to do here:

a. Check the boxes for the models you want available

Don’t surface every model the provider offers. Most providers list 30+ variants — some old, some specialized, some experimental. Pick two or three that cover your needs:

• One latest-generation model (gpt-4o, claude-3-5-sonnet, or whatever’s current) — your default for most work.
• One smaller/cheaper model (gpt-4o-mini, claude-3-haiku) — good for simple tasks where you’d rather save cost.
• (Optional) one specialized model if you have a specific need (e.g., a long-context model for document review).

Checked models appear in the model picker for chats and agents. Unchecked models don’t, even though your account has access to them.

b. Set a default model

Pick which model is the default when this provider is selected. Most teams set this to their main latest-generation model. The default is what an agent uses if it doesn’t have a more specific preference set.

There’s no Save button — auto-save is on. Your changes apply immediately.

Tip: You can come back to this page anytime to add or remove models as the provider releases new ones. Click Refresh from provider to re-fetch the available list.

Step 6 (optional): Set up “Auto” routing

Above the providers list, you’ll see an Auto routing section. Helper: “When a chat thread or agent uses Auto for the model, Auxot starts from the provider (and optional model) you set here, then applies that provider’s fallback rules if needed.”

Plain English: when a user (or agent) picks “Auto” from the model selector, Auxot uses the provider you set here as its first try. If you don’t set anything, Auxot uses a built-in priority order (GPU → CLI → OpenAI → Anthropic — whichever is connected first).

For a brand-new account with one cloud provider, you can leave Auto routing alone. Come back to it once you have multiple providers and want a preference.

Tip: Even with one provider connected, “Auto” still works — it just always picks that provider. The Auto setting becomes more useful when you have a primary and a fallback (e.g., “prefer Anthropic, fall back to OpenAI if Anthropic is down or over quota”). You configure those fallback rules on each provider’s detail page.

What’s next

• → Tutorial 10: Bootstrap a GPU worker — if you want AI running on your own hardware so no data leaves your building, this is the path.
• → Tutorial 03: Take Auxot’s pulse in 10 seconds — System Health shows your providers’ status live; useful right after connecting one.
• → Tutorial 07: Create an agent from scratch — now that the AI model is connected, build the agents that use it.

Reference

• Pages in Auxot: Settings → Providers
• Supported provider types: OpenAI, Anthropic, GPU (Tutorial 10), CLI (Tutorial 10)
• Required: API key from your provider account, payment method on the provider’s side
• Permissions: org admins create/edit providers; all users can see what’s connected
• Free tier limit: one provider of each type
• Auto routing: built-in priority is GPU → CLI → OpenAI → Anthropic; override with org default in Auto routing section
• See also: Tutorial 03: Take Auxot’s pulse, Tutorial 07: Create an agent from scratch, Tutorial 10: Bootstrap a GPU worker