AI providers (bring your own key)
Connect your AI provider key. Required for every AI feature.
How AI works in Fluxo Flow
Fluxo Flow is bring-your-own-key. Your workspace connects its own AI provider account (such as OpenAI, Anthropic or Google), and every AI call made for that workspace uses that key. Your provider bills you directly for tokens at their normal rates. Fluxo Flow has no AI credits, and never marks up or resells AI usage. Your Fluxo Flow plan covers the platform: workflows, seats, schedules and integrations.
| Feature | Needs an AI key? |
|---|---|
| Agent, Observer and AI Action steps (real Run, schedules, webhooks) | Yes |
| Marketplace agents | Yes |
| Canvas insights and AI-generated node tips | Yes |
| Support assistant chat | Yes (or open a ticket for a human) |
| Dry Run | No — AI steps are simulated |
| Triggers, integrations, routers, transformers, validators, outputs, human review | No |
| Curated tips, docs, analytics, run history | No |
Supported providers
| Provider | What you need | Where to get it |
|---|---|---|
| OpenAI | API key (sk-…) | platform.openai.com/api-keys |
| Anthropic | API key (sk-ant-…) | console.anthropic.com/settings/keys |
| Google Gemini | API key (AIza…) | aistudio.google.com/apikey |
| Mistral | API key | console.mistral.ai/api-keys |
| Groq | API key (gsk_…) | console.groq.com/keys |
| xAI (Grok) | API key (xai-…) | console.x.ai |
| DeepSeek | API key | platform.deepseek.com/api_keys |
| OpenRouter | API key (sk-or-…) — one key for hundreds of models | openrouter.ai/settings/keys |
| Together AI | API key | api.together.ai/settings/api-keys |
| Fireworks AI | API key (fw_…) | fireworks.ai/account/api-keys |
| Azure OpenAI | Endpoint, key, and your deployment names | Azure portal → your Azure OpenAI resource → Keys and Endpoint |
| AWS Bedrock | Region, access key ID and secret access key | AWS IAM → Users → Security credentials |
| Google Vertex AI | Service account key (JSON) and location | Google Cloud → IAM → Service accounts |
| Custom | Base URL of any OpenAI-compatible service, plus a key and a default model | Your provider's docs |
| Local model | Tunnel URL to Ollama, LM Studio, llama.cpp or vLLM on your own machine, plus the model name | See "Local models" below |
Local models (Ollama, LM Studio)
Run a model on your own computer or server and use it in any AI step, at no per-token cost. Fluxo runs in the cloud, so it can't reach localhost: start a tunnel to your local server (for Ollama: cloudflared tunnel --url http://localhost:11434 --http-host-header localhost:11434; LM Studio uses port 1234), then in Settings → AI Providers → Local model paste the https address it prints and pick your model. Your computer and the tunnel must be running whenever a workflow uses the model, including scheduled runs. Quick tunnels get a new address each time; a named Cloudflare tunnel keeps one. Keep the address private.
Use a dedicated key
Create a key just for Fluxo Flow (for example, named "fluxo-production"). You can then set a spending limit on it with your provider and revoke it without affecting anything else. Make sure the account has billing or credit set up — keys on accounts without credit are rejected with an "out of credit" error.
Azure, AWS Bedrock and Vertex AI
Azure OpenAI
- In Azure AI Foundry, deploy the models you want (for example gpt-4o-mini) and note each deployment's name.
- In the Azure portal, open your Azure OpenAI resource → Keys and Endpoint. Copy the endpoint and Key 1.
- In Fluxo Flow, connect Azure OpenAI with the endpoint, key and your deployment names (comma-separated). The first deployment is the default for insights and tips.
AWS Bedrock
- In the Bedrock console, open Model access and enable the models you want in your region.
- Create an IAM user (or use an existing one) with permission to call Bedrock (bedrock:InvokeModel, bedrock:InvokeModelWithResponseStream and bedrock:ListFoundationModels), and create an access key for it.
- In Fluxo Flow, connect AWS Bedrock with the region, access key ID and secret access key. Optionally set a default model ID such as amazon.nova-lite-v1:0.
Google Vertex AI
- In your Google Cloud project, enable the Vertex AI API.
- Create a service account with the Vertex AI User role, then create a JSON key for it and download it.
- In Fluxo Flow, connect Google Vertex AI and paste the whole JSON file. Set the location if you don't use us-central1.
Custom endpoints must be public
A custom base URL (or Azure endpoint) must be a public https address. Local or private addresses (like localhost or 10.x.x.x) are blocked for security, so self-hosted models need a public, secured URL.
Connect it to your workspace
- Open Settings → AI Providers. Owners and admins can connect, replace, test and disconnect keys; everyone else can see which providers are connected.
- Find your provider (OpenAI, Anthropic, Google Gemini, Mistral, Groq, xAI, DeepSeek, OpenRouter, Together AI, Fireworks AI, Azure OpenAI, AWS Bedrock, Google Vertex AI, a custom OpenAI-compatible endpoint, or a local model), fill in its fields and press Connect.
- Fluxo Flow checks the key with the provider before saving it. If the provider rejects it, nothing is saved and you'll see why.
- Once connected, the card shows that it's connected and a short hint of the key (its first and last few characters).
- To rotate a key, press Replace and enter the new one. Press Test to re-check a saved key. To stop all AI use for that provider, press Disconnect.
Choosing models
Each Agent, Observer and AI Action step has a Model setting listing models from the providers you've connected. Press "Browse all models or enter a model ID" to load a provider's full live model list, search it, or type any model ID the provider supports. The provider of the model decides which credentials are used, and a real Run is blocked (with a message pointing here) if any AI step uses a provider that isn't connected.
| Model | Provider | Good for | Relative cost |
|---|---|---|---|
| GPT-4o mini | OpenAI | Default for new AI steps; fast classification and extraction | $ |
| GPT-4o | OpenAI | General drafting and reasoning | $$ |
| GPT-4 Turbo | OpenAI | Older strong model, kept for existing workflows | $$$ |
| Claude Haiku 4.5 | Anthropic | Fast, low-cost, high-volume work | $ |
| Claude Sonnet 5.5 | Anthropic | Balanced drafting, summarising, multi-step reasoning | $$ |
| Claude Opus 5.5 | Anthropic | Hardest problems; slower and pricier | $$$ |
Costs and limits
- Token costs are billed by your AI provider to the account that owns the key, at their published prices.
- Each run shows an estimated AI cost, so you can see which steps are expensive. Costs use list prices for popular models; for other models they're a rough estimate — your provider's bill is the source of truth.
- Every server run stops if its estimated AI cost passes $2, to protect you from a runaway loop. Observer steps can also set their own maximum cost.
- For a hard ceiling, set a monthly spending limit on the key with your provider.
How your key is protected
- Keys are encrypted at rest (pgcrypto) with a server-side key the database never stores.
- A key is only ever used for AI calls made for its own workspace.
- The full key is never sent back to a browser. Settings shows only a short hint such as sk-…1234.
- Adding, replacing, testing and removing keys is recorded in Settings → Activity log (owners and admins can see it). The key itself never is.
Troubleshooting
| Message | What it means | Fix |
|---|---|---|
| No AI provider connected | The workspace has no key for that model's provider. | Connect a key in Settings → AI Providers, or pick a model from a provider you've connected. |
| Your AI provider key was rejected | The key is wrong, revoked or disabled. | Create a new key with the provider and press Replace in Settings → AI Providers. |
| Your AI provider account is out of credit | The provider account has no billing or has used its credit. | Add billing or credit with the provider, then press Test in Settings → AI Providers. |
| Your AI provider is rate-limiting requests | Too many requests on that key right now. | Wait a few seconds and retry. Raise your usage tier with the provider if it keeps happening. |
| The AI service had an error | The provider had an outage or a temporary error. | Retry shortly and check the provider's status page. |
| Claude declined this request | The model refused the prompt. | Rephrase the prompt or pick a different model. |
You'll be told when a key breaks
If a scheduled or webhook run finds your key rejected or out of credit, owners and admins get a notification in the bell.
Related guides
Last updated . New to Fluxo Flow? See what it does and how pricing works.