AI Provider Reference
AeroFTP v4.2.7 defines 26 AI provider types, including custom endpoints. Most adapters use an OpenAI-compatible API; Anthropic and Gemini have native formats. A configured provider type does not guarantee that every model supports tools, vision or thinking.
Provider Overview
| # | Provider | Type | Base URL | Auth |
|---|---|---|---|---|
| 1 | OpenAI | openai | https://api.openai.com/v1 | API Key |
| 2 | Anthropic | anthropic | https://api.anthropic.com/v1 | API Key |
| 3 | Google Gemini | google | https://generativelanguage.googleapis.com/v1beta | API Key |
| 4 | xAI (Grok) | xai | https://api.x.ai/v1 | API Key |
| 5 | OpenRouter | openrouter | https://openrouter.ai/api/v1 | API Key |
| 6 | Ollama (Local) | ollama | http://localhost:11434 | None |
| 7 | Kimi (Moonshot) | kimi | https://api.moonshot.cn/v1 | API Key |
| 8 | Qwen (Alibaba) | qwen | https://dashscope-intl.aliyuncs.com/compatible-mode/v1 | API Key |
| 9 | DeepSeek | deepseek | https://api.deepseek.com | API Key |
| 10 | Mistral | mistral | https://api.mistral.ai/v1 | API Key |
| 11 | Groq | groq | https://api.groq.com/openai/v1 | API Key |
| 12 | Perplexity | perplexity | https://api.perplexity.ai | API Key |
| 13 | Cohere | cohere | https://api.cohere.com/compatibility/v1 | API Key |
| 14 | Together AI | together | https://api.together.xyz/v1 | API Key |
| 15 | AI21 Labs | ai21 | https://api.ai21.com/studio/v1 | API Key |
| 16 | Cerebras | cerebras | https://api.cerebras.ai/v1 | API Key |
| 17 | SambaNova | sambanova | https://api.sambanova.ai/v1 | API Key |
| 18 | Fireworks AI | fireworks | https://api.fireworks.ai/inference/v1 | API Key |
| 19 | NVIDIA NIM | nvidia | https://integrate.api.nvidia.com/v1 | API Key |
| 20 | Z.AI | zai | https://api.z.ai/api/paas/v4 | API Key |
| 21 | Hyperbolic | hyperbolic | https://api.hyperbolic.xyz/v1 | API Key |
| 22 | Novita | novita | https://api.novita.ai/v3/openai | API Key |
| 23 | Yi | yi | https://api.lingyiwanwu.com/v1 | API Key |
| 24 | Kilo Gateway | kilo | https://api.kilo.ai/api/gateway | API Key |
| 25 | Amazon Bedrock | bedrock | https://bedrock-runtime.<region>.amazonaws.com/openai/v1 | Bedrock API key |
| 26 | Custom | custom | User-defined | User-defined |
For Amazon Bedrock, replace <region> with your AWS Region and keep the /openai/v1 suffix, for example https://bedrock-runtime.eu-north-1.amazonaws.com/openai/v1 (the shipped preset). Use a Bedrock API key and a model/API supported in that Region. See the official AWS endpoint guidance; this is not an OpenAI API key or the plain regional host without its API path.
Model capabilities
Tools, vision, thinking, structured output and sampling controls depend on the selected model and adapter, rather than the provider name alone. AI Settings tracks capability discovery, explicit overrides and their source. Use the model's current capability record and test the operation you need. A native API feature listed in model metadata is not a promise that AeroAgent has implemented that feature.
Models
AeroFTP does not ship with hardcoded default models. Use the Models button in AI Settings to fetch available models from each provider dynamically.
For Ollama, the Detect button queries GET /api/tags to list locally installed models. You can also pull new models directly from the UI via POST /api/pull with streaming progress.
Provider-Specific Notes
Anthropic
- Uses native Anthropic API format (not OpenAI-compatible)
- Prompt caching: Supports the adapter's
cache_controlmechanism; billing and eligibility depend on the provider and model - System prompt: Sent as top-level
systemfield - Thinking: Native
thinkingblocks with configurable budget (0-100K tokens)
Google Gemini
- Uses Gemini-native API format
- System instruction: Sent as top-level
system_instructionfield (not as a message) - Code execution: Supports
executableCode/codeExecutionResultblocks - Context caching:
gemini_create_cachefor reusing large context prefixes
Ollama
- Runs locally, no API key required
- Model families: 8 detected profiles (llama, mistral, phi, gemma, qwen, deepseek, codellama, vicuna) with family-specific prompt styles
- GPU monitoring:
ollama_list_runningshows active models and VRAM usage - Model pull: Download models from the Ollama registry with streaming progress bar
OpenRouter
- Aggregator: routes to 100+ models from multiple providers
- Uses OpenAI-compatible format
- Structured output support depends on the underlying model
DeepSeek
Endpoint Format
DeepSeek's base URL is https://api.deepseek.com without a trailing /v1. The /v1/chat/completions path is appended by the client.
Cohere
Compatibility Endpoint
Cohere uses the compatibility endpoint at https://api.cohere.com/compatibility/v1, not the native /v2 API. Tool format is text-based, not native function calling.
stream_options
The stream_options: { include_usage: true } field is included in streaming requests to receive token usage in the final chunk. This field is excluded for Cohere and Perplexity, which reject unknown fields.
Provider Profiles
Each provider has a built-in personality profile that optimizes the system prompt for that provider's strengths:
- Parameter presets: Creative, Balanced, and Precise temperature/top_p configurations
- Capability awareness: The system prompt adapts based on whether the provider supports tools, vision, and thinking
- Ollama model-specific templates: Prompt format adapts to the detected model family
Thinking Budget
For providers that support reasoning/thinking (Anthropic, OpenAI o3, Gemini, DeepSeek), AeroAgent offers 5 presets:
| Preset | Token Budget | Use Case |
|---|---|---|
| Off | 0 | Simple queries |
| Light | 1,024 | Quick reasoning |
| Balanced | 8,192 | General use |
| Deep | 32,768 | Complex analysis |
| Maximum | 100,000 | Full reasoning chain |
A range slider allows fine-tuning between 0 and 100,000 tokens.
Adding a Custom Provider
Any OpenAI-compatible API can be added as a Custom provider:
- Go to Settings > AeroAgent > Providers
- Click Add Provider or use the Provider Marketplace
- Select Custom
- Enter the base URL and API key
- Use the Models button to fetch available models
The base URL should point to the root of the API (e.g., https://my-provider.com/v1). AeroFTP appends /chat/completions automatically.