AI Providers

FleetRun supports 12+ AI model providers. All cloud providers use the OpenAI-compatible API format except Anthropic, which uses its native SDK. Add any provider in Settings → Providers.

How providers work

Each agent has a model_provider and model_name field. When a task runs, the agent worker resolves the provider from the providers table, reads the API key from the corresponding environment variable, and calls the model.

If the primary provider fails with a network error and ANTHROPIC_API_KEY is set, the router automatically falls back to Anthropic Claude.

Adding a provider

  1. Go to Settings → Providers in the FleetRun app.
  2. Click Add provider and select from the quickstart list, or enter custom values.
  3. Set the corresponding environment variable in your Vercel project settings.
  4. The provider health indicator turns green once the key is valid and the endpoint is reachable.

Provider reference

Anthropicanthropic
Base URLhttps://api.anthropic.com
Env varANTHROPIC_API_KEY
Get keyconsole.anthropic.com
Modelsclaude-opus-4-8claude-sonnet-4-6claude-haiku-4-5-20251001

Supports extended thinking and streaming. Native SDK used (not OpenAI-compat).

OpenAIopenai
Base URLhttps://api.openai.com/v1
Env varOPENAI_API_KEY
Get keyplatform.openai.com/api-keys
Modelsgpt-4ogpt-4o-minio3o4-minigpt-4.1

OpenAI-compatible endpoint. Supports function calling and streaming.

Google Geminigemini
Base URLhttps://generativelanguage.googleapis.com/v1beta/openai
Env varGEMINI_API_KEY
Get keyaistudio.google.com/app/apikey
Modelsgemini-2.5-progemini-2.5-flashgemini-2.0-flashgemini-1.5-pro

Uses the OpenAI-compatible endpoint exposed by Google AI Studio.

Groqgroq
Base URLhttps://api.groq.com/openai/v1
Env varGROQ_API_KEY
Get keyconsole.groq.com
Modelsllama-3.3-70b-versatilellama-3.1-8b-instantmixtral-8x7b-32768

LPU inference — fastest available latency for Llama and Mixtral models.

Mistralmistral
Base URLhttps://api.mistral.ai/v1
Env varMISTRAL_API_KEY
Get keyconsole.mistral.ai
Modelsmistral-large-latestmistral-small-latestcodestral-latestmistral-nemo

Codestral is optimized for code completion and generation tasks.

DeepSeekdeepseek
Base URLhttps://api.deepseek.com/v1
Env varDEEPSEEK_API_KEY
Get keyplatform.deepseek.com
Modelsdeepseek-chatdeepseek-reasoner

deepseek-reasoner uses chain-of-thought and is ideal for multi-step reasoning.

xAI Grokxai
Base URLhttps://api.x.ai/v1
Env varXAI_API_KEY
Get keyconsole.x.ai
Modelsgrok-3grok-3-minigrok-2-1212

Grok 3 has real-time web access. OpenAI-compatible API.

Together AItogether
Base URLhttps://api.together.xyz/v1
Env varTOGETHER_API_KEY
Get keyapi.together.xyz/settings/api-keys
Modelsmeta-llama/Llama-3.3-70B-Instruct-Turbodeepseek-ai/DeepSeek-V3Qwen/Qwen2.5-72B-Instruct-Turbo

200+ open-source models on a GPU cloud. Good balance of cost and quality.

Perplexityperplexity
Base URLhttps://api.perplexity.ai
Env varPERPLEXITY_API_KEY
Get keyperplexity.ai/settings/api
Modelssonar-prosonarsonar-reasoning-prosonar-reasoning

Sonar models include live web search. Ideal for research agents.

Fireworksfireworks
Base URLhttps://api.fireworks.ai/inference/v1
Env varFIREWORKS_API_KEY
Get keyfireworks.ai/account/api-keys
Modelsaccounts/fireworks/models/llama-v3p3-70b-instructaccounts/fireworks/models/deepseek-v3

Model IDs use the accounts/fireworks/models/<name> format.

OpenRouteropenrouter
Base URLhttps://openrouter.ai/api/v1
Env varOPENROUTER_API_KEY
Get keyopenrouter.ai/keys
Modelsanthropic/claude-opus-4google/gemini-2.5-prodeepseek/deepseek-r1

Single key routes to 200+ models. Prefix model names with the provider slug.

Ollama (local)ollama_nas
Base URLhttp://10.1.1.20:11434/v1
Env varOLLAMA_NAS_URL
Get keyollama.com/download
Modelsgemma4:26bllama3.3:70bqwen2.5:14bqwen2.5-coder:7b

Runs locally on your NAS. No API key required. Model IDs are Ollama tag names.

Custom OpenAI-compatible endpoints

Any endpoint that speaks the OpenAI chat completions API can be added as a custom provider. In Settings → Providers, click Add provider and fill in:

  • Key — a unique slug (e.g. my_provider)
  • Label — display name
  • Base URL — endpoint ending in /v1
  • API key env var — the env var name that holds the key (optional for local)
  • Tagcloud or local