57 AI Providers, One API, Zero Markup.
Every AI provider AeionClaw can route to. Switch between OpenAI and Anthropic for reasoning. Use Gemini Flash for bulk classification. Use Groq for sub-100ms latency. Use ElevenLabs for voice. Use Tripo3D for 3D meshes. Use Voyage for embeddings. Or use your own self-hosted Ollama / vLLM. The router picks the right provider for each task — and you pay the provider's published rate with zero Aeion markup.
How the Provider Router Works
AeionClaw's Aeion AI service layer is a unified gateway across 57 providers. Configure once; switch routing strategy per-task or per-skill.
Text / Reasoning Providers (19)
OpenAI
GPT-4o, GPT-4.5, o1, o3, o4, function calling, vision
Anthropic
Claude 4 Opus, 4 Sonnet, 3.7 Sonnet, Haiku, vision, tool use
Gemini 2.5 Pro, 2.5 Flash, 2.0 Flash, vision, function calling
Groq
Ultra-fast Llama 3.1 70B / Mixtral / Gemma at sub-100ms latency
DeepSeek
R1 / R2 reasoning models, cost-effective at scale
Mistral
Mistral Large, Codestral, Mistral 7B (self-hostable)
Perplexity
Online-grounded LLM with citations
xAI
Grok 2, Grok 3, real-time X integration
Azure OpenAI
Enterprise-deployed OpenAI models
Together / Fireworks
Open-source LLM serving
OpenRouter
Meta-router across 100+ open models
Ollama
Self-hosted local inference (any GGUF model)
HuggingFace
Transformers via Inference Endpoints + TGI
Moonshot (Kimi)
Chinese-language + 200K context window
Cerebras
Inference on wafer-scale chips, sub-100ms latency
Blackbox
Code-specialized LLM
Novita / Bedrock
Specialized inference endpoints
Image Generation Providers (5)
DALL-E 3 (OpenAI)
High-quality general-purpose image generation
Stability AI
Stable Diffusion 3.5, SD Ultra, ControlNet support
Ideogram 2.0
Best-in-class text rendering inside images
Recraft V3
Vector + raster, design-focused output
fal.ai
Hosted SD + Flux + custom-model serving
3D Generation Providers (3)
Tripo3D
Text-to-3D mesh, high topology + UV unwrapping
Meshy AI
PBR-textured meshes, animation-ready rigging
Luma
DreamMachine for video, Photon for image, 3D capture
TTS / Voice Synthesis Providers (16)
ElevenLabs
Sona, Multilingual v2, voice-clone support, 30+ languages
Cartesia Sonic
Real-time TTS with sub-100ms latency
Kokoro TTS
Self-hostable TTS, brand-voice-aware
PlayHT
Studio-quality voice models
LMNT
Real-time conversational TTS
AssemblyAI / Rev.ai
Multi-purpose audio platforms
Resemble AI
Voice cloning + emotion control
Rime AI
Hyper-realistic conversational TTS
Speechmatics
Multilingual + speaker identification
PocketTTS
Localhost (`baseUrl: "http://localhost:8001"`)
DashScope (Qwen3-TTS)
Alibaba's open-source TTS
Cohere Transcribe
Multi-language audio platform
Voxtral / Supertonic / Novita
Specialized TTS providers
STT / Speech-to-Text Providers (9)
AssemblyAI
High-accuracy transcription + speaker diarization
Deepgram
Real-time streaming transcription, low latency
Gladia
Speaker identification + sentiment + multilingual
Voxtral STT
Specialized transcription provider
Speechmatics
50+ languages, real-time + batch
Rev.ai
Human-grade transcription, 14+ languages
Whisper (OpenAI)
Open-source speech recognition
Cohere Transcribe
Multi-purpose audio platform
LlamaParse
Document + audio parsing
Video Generation Providers (4)
Runway (Gen-3)
Text-to-video, motion editing, professional output
Video Intelligence
Specialized video AI services
Supertonic
Real-time video synthesis
Luma Photon
High-resolution video generation
Embeddings + Reranking Providers (6)
Cohere
embed-v3 + embed-multilingual + Cohere Rerank
Voyage AI
voyage-3 + voyage-3-large + voyage-3-lite
Jina AI
jina-embeddings-v3, multilingual support
Nomic AI
Nomic Embed Text + Nomic Vision
Mixedbread AI
Open-source SOTA embeddings
OpenAI
text-embedding-3-small + 3-large
Specialty Providers (2 + custom)
Voice Cloning
Resemble AI + ElevenLabs (consent + license tracking)
Document Parsing
LlamaParse (PDFs, images, complex tables)
OpenRouter
Meta-router across 100+ models for niche use cases
Bring Your Own Model (BYOM)
Custom adapters via `AIProvider` interface for any OpenAI-compatible endpoint (Ollama, vLLM, custom inference servers, internal model deployments, fine-tuned models)
Provider Cost Cheat Sheet
Typical pricing as of May 2026 (per 1M tokens, input/output for text models).
Provider questions
Yes — Bring Your Own Model (BYOM) via custom adapters. Provide an OpenAI-compatible baseURL + the `AIProvider` interface implementation. Self-hosted Ollama, vLLM, HuggingFace TGI, internal model servers all supported. Air-gapped Enterprise deployments use BYOM exclusively.
Zero. AeionClaw is included free on every Aeion deployment — no tier gates provider count, retention, or cost caps (see /plans). The AI calls cost exactly what each provider charges — no per-token uplift, no percentage on spend.
Daily auto-refresh pulls new pricing within 24 hours. Per-tenant cost caps protect you from surprise overruns. Most price increases are pre-announced; you can adjust caps before the change.
Yes — per-tenant + per-role provider allowlists. Some regulated customers restrict to "only providers with US data residency" or "only providers with HIPAA BAAs."
Circuit breaker pattern. The router auto-routes to next-best provider matching the strategy. Unhealthy provider gets backoff timer. Recovery automatic when health restores.
By moving bulk-classification tasks (translation, summarization, simple Q&A) from GPT-4o to Gemini Flash. Real customer cut spend 99.5% on background tasks ($1,350 → $6.90/month). High-reasoning tasks stay on GPT-4o or Claude.
Yes — per-step routing. A multi-step workflow might use Groq for fast classification, Claude for nuanced reasoning, OpenAI for tool use, and ElevenLabs for voice output — all in one user-facing conversation.
Aeion doesn't fine-tune. If you fine-tune with OpenAI / Anthropic / etc., bring the fine-tuned model ID via BYOM. Your data stays in your tenant.
Yes — multimodal models (GPT-4o, Claude vision, Gemini Pro vision) route through the same gateway. Single API call across modalities.