LLM Providers
Every thinking node in a flow calls an AI provider, and every provider wants an API key. Flowera covers the common case for you — but you can bring your own key whenever you'd rather.
Bringing your own model keys
Two reasons to add your own:
- Billing. Usage on your key bills to your provider account instead of your Flowera wallet. If you already have negotiated pricing or committed spend with a provider, use it.
- Coverage. Your key reaches models the platform key doesn't offer — a preview model, a fine-tune, a region-restricted deployment.
If neither applies, don't bother. The platform key works, and one fewer secret to rotate is worth something.
Supported providers
Flowera has credential types for the major providers and most of the smaller ones:
| Group | Providers |
|---|---|
| Mainstream | OpenAI, Anthropic, Google (Generative AI and Vertex), Azure OpenAI, Mistral, Cohere |
| Fast inference | Groq, Cerebras, Fireworks, Together, DeepSeek, NVIDIA NGC |
| Aggregators | OpenRouter, LiteLLM |
| Self-hosted | Ollama, LocalAI |
| Other | HuggingFace, Perplexity, IBM watsonx, Baidu Qianfan, Alibaba |
Which models each offers is a separate question — see Choosing a model.
It's the exception on this page. You point the node at your server's base URL and that's the whole configuration — there is nothing to create under Credentials for it.
Adding a provider key
- Open Credentials, stay on the API Credentials tab, and press Add Credential.
- Pick the provider's credential type (for example OpenAI API).
- Give it a Credential Name you'll recognise in a dropdown, and paste the key.
- Save.
The key is encrypted immediately and never shown again. To change it, edit the credential and paste a new one.
When Flowera's default key is used
Some providers are marked System in the credentials list — these are Flowera's own platform keys, and they're what makes a brand-new workspace able to run a flow without any setup at all.
| Situation | What gets used |
|---|---|
| You pick a model and attach no credential | The platform key, if that provider has one |
| You attach your own credential to the node | Your key |
| The provider has no platform key and you attach none | Nothing — the node fails with a "Model is required" style error |
Usage on a platform key is billed to your Flowera wallet at the rates shown on the Pricing tab. Usage on your own key is billed by the provider, and doesn't appear in your Flowera spend at all — worth remembering when the two dashboards disagree.
Azure OpenAI
Azure OpenAI has a platform key too, so Azure ChatOpenAI and Azure OpenAI Embeddings work without an Azure subscription of your own. The Azure nodes have their own rates — see the Pricing tab for the current numbers.
The Azure model list is shorter. Azure stops accepting new deployments of a model once it
enters its deprecation window, which currently rules out the GPT-4.1 and GPT-4o families as well
as o1, o3-mini and o4-mini — all still available on the plain OpenAI nodes. Azure offers the
GPT-5 family and o3. If a model you want is missing from the Azure dropdown, that is why; use the
OpenAI node for it.
One thing behaves differently on Azure: requests are routed by deployment name, not by the model name in the request. Flowera names each deployment after the model it serves, so selecting a model in the node is all you need to do.
If you attach your own Azure credential, two fields can change that:
- Deployment name on the credential — optional. Setting it pins every model on that credential to that one deployment, so the node's model dropdown no longer changes where the request goes. Leave it empty if your deployments are named after their models.
- Deployment Name on the node (under additional parameters) — overrides both, for one node only. Use it when a single node needs a deployment that doesn't match its model name.
The node prefers them in that order: node field, then credential, then the selected model.
On the shared Flowera credential, Deployment Name and BasePath are ignored — the request goes to the deployment matching the model you selected, at Flowera's own endpoint. Flowera's key must reach Flowera's Azure resource and nowhere else, and you are billed for the model the node shows. Both fields work normally the moment you connect your own credential.
If your credential already has a deployment name filled in, it keeps winning — your existing flows carry on hitting the same deployment they always have. To start routing by model instead, clear that field on the credential.
Choosing a key per node
Credentials are picked per node, not per flow or per workspace. That's more useful than it sounds:
- Run the expensive reasoning node on your own negotiated-rate key, and leave cheap routing nodes on the platform key.
- Give a client's flow the client's own key, so their usage lands on their bill.
- Test a new provider on one node before moving a whole flow to it.
See Models for how a node picks its model.
Troubleshooting
| Symptom | Likely cause |
|---|---|
| "Model is required" after saving a node | No credential attached and no platform key for that provider — pick the provider and model again |
401 / invalid API key | The key was revoked or mistyped; paste a fresh one |
429 / rate limited | Your provider account's own limit, not Flowera's — raise it with the provider |
| Works on one node, fails on another | The other node has a different credential (or none) attached |
DeploymentNotFound on Azure | Your Azure resource has no deployment named after the selected model — create one, or set Deployment Name on the node |
| Spend on your provider bill but not in Flowera | Expected: your own key bills there, not here |
Related
- Choosing a model — which model for which job
- Credentials overview — how credentials are stored and scoped
- Pricing — what platform-key usage costs