☁️ API Providers
OpenRouter
API★★★★★
Universal API gateway to 200+ models from every major provider. One API key gives access to everything — Claude, GPT-4, Gemini, DeepSeek, Llama, Mistral, and more. The recommended starting point for most users.
Models200+ models — Anthropic, OpenAI, Google, Meta, Mistral, DeepSeek, and dozens more
PricingPay-per-token at provider rates. No monthly fee. Free tier models available.
Best forBest default provider. One key, all models. Easy switching between providers.
Free tier✅ Free tier
Config varsOPENROUTER_API_KEY
Pros- One key for 200+ models
- Free tier available
- Easy model switching
- No monthly fee
- Credential pooling supported
- Rate limit fallback
Cons- Slight latency overhead vs direct API
- Some niche models not available
- Dependent on OpenRouter uptime
Setup guideSign up at openrouter.ai → create API key → `echo 'OPENROUTER_API_KEY="sk-or-v1-..."' >> ~/.hermes/.env` → `hermes model` → pick a model → start chatting
☁️ API Providers
Anthropic
API★★★★★
Claude models — Sonnet, Haiku, and Opus. Best-in-class for coding, complex reasoning, and nuanced instruction following. Hermes's recommended premium model.
ModelsClaude Sonnet, Haiku, and Opus (current generations)
PricingPay-per-token - see anthropic.com/pricing for current rates
Best forComplex coding, deep reasoning, nuanced tasks. The smartest model for hard problems.
Free tierNo — paid API only
Config varsANTHROPIC_API_KEY
Pros- Best coding & reasoning model
- Excellent instruction following
- Large context window (200K)
- Reliable API
- Good safety guardrails
Cons- Expensive (Opus especially)
- No free tier
- Rate limits on lower tiers
- Vision quality varies
Setup guideSign up at anthropic.com → create API key → `echo 'ANTHROPIC_API_KEY="sk-ant-..."' >> ~/.hermes/.env` → `hermes config set model.default "claude-sonnet-4-20250514"` → `hermes config set model.provider "anthropic"`
☁️ API Providers
OpenAI
API★★★★☆
OpenAI GPT models through the openai-api provider. Broad capabilities with excellent tool-use performance. The original LLM API.
ModelsCurrent OpenAI GPT and o-series models (see platform.openai.com)
PricingGPT-4o: ~$10/1M tokens. GPT-4o-mini: ~$1/1M. o-series: $15-60/1M.
Best forGeneral purpose. Good tool-use, broad knowledge, fast responses with mini models.
Free tierNo — paid API only
Config varsOPENAI_API_KEY
Pros- Excellent tool-use performance
- Fast mini models (cheap)
- Broad model selection
- Reliable infrastructure
- Great documentation
Cons- No free API tier
- o-series models are expensive
- Usage limits on new accounts
- Less nuanced than Claude for reasoning
Setup guideSign up at platform.openai.com → create API key → `echo 'OPENAI_API_KEY="sk-..."' >> ~/.hermes/.env` → `hermes config set model.default "gpt-4o"` → `hermes config set model.provider "openai-api"`
☁️ API Providers
DeepSeek
API★★★★★
DeepSeek Chat and DeepSeek Reasoner. Excellent quality-to-price ratio. Strong coding performance comparable to Claude at a fraction of the cost.
ModelsDeepSeek Chat and DeepSeek Reasoner (current generations)
PricingChat: ~$0.27/1M tokens. Reasoner: ~$0.55/1M. Fraction of premium providers.
Best forBest value provider. Near-premium quality at budget prices. Excellent for coding.
Free tier✅ Free tier
Config varsDEEPSEEK_API_KEY
Pros- Excellent price/performance
- Strong coding ability
- Free tier available
- Reasoner model for complex tasks
- Chinese language support
Cons- Fewer models than OpenAI/Anthropic
- API reliability varies
- Less ecosystem/support
- Some rate limits on free tier
Setup guideSign up at platform.deepseek.com → create API key → `echo 'DEEPSEEK_API_KEY="sk-..."' >> ~/.hermes/.env` → `hermes config set model.default "deepseek-chat"` → `hermes config set model.provider "deepseek"`
☁️ API Providers
Google Gemini
API★★★★☆
Gemini models including Gemini 3 Flash. Very fast, very cheap, huge context window (1M+ tokens). Great for processing large documents.
ModelsGemini 3 Flash and other current Gemini models
PricingFlash: ~$0.10/1M tokens (one of the cheapest). Pro: ~$1.25/1M.
Best forCost-effective general use. Processing large documents (1M+ token context). Fast responses.
Free tier✅ Free tier
Config varsGOOGLE_API_KEY, GEMINI_API_KEY
Pros- Very cheap (Flash)
- 1M+ token context window
- Fast response times
- Free tier available
- Google ecosystem integration
Cons- Less consistent than Claude for coding
- Fewer third-party tools
- Experimental models can be unstable
- Some features lag behind OpenAI
Setup guideGet API key at aistudio.google.com → `echo 'GOOGLE_API_KEY="AIza..."' >> ~/.hermes/.env` → `hermes config set model.default "gemini-3-flash"` → `hermes config set model.provider "gemini"`
☁️ API Providers
xAI / Grok
API★★★☆☆
xAI's Grok models. Good general performance with real-time knowledge. Integrated with X/Twitter ecosystem.
ModelsCurrent Grok models (see x.ai); Grok OAuth is available for SuperGrok and X Premium+ subscribers
PricingPay-per-token - see x.ai for current rates
Best forX/Twitter integration. Real-time knowledge. Image generation via Grok-Imagine.
Free tierLimited — free tier for X Premium subscribers
Config varsXAI_API_KEY
Pros- Real-time knowledge
- X/Twitter integration
- Grok-Imagine image gen
- Competitive pricing
Cons- Newer provider, less mature
- Fewer models
- Smaller ecosystem
- Limited free options
Setup guideSign up at x.ai → create API key → `echo 'XAI_API_KEY="xai-..."' >> ~/.hermes/.env` → `hermes config set model.default "grok-3"` → `hermes config set model.provider "xai"`
🔑 OAuth / Platform
GitHub Copilot
OAuth★★★★☆
GitHub Copilot as LLM provider. Uses Copilot's backend models (currently based on GPT-4o and Claude). Free for verified students, teachers, and OSS maintainers.
ModelsCopilot models (GPT-4o based, Claude Sonnet)
Pricing$10/month (Pro) or included in GitHub Enterprise. Free for students/teachers/OSS.
Best forFree/cheap access to premium models. Perfect if you already have Copilot.
Free tier✅ Free tier
Config varsCOPILOT_GITHUB_TOKEN or hermes login
Pros- Free for students/teachers/OSS
- Good model quality
- No separate billing
- OAuth setup is smooth
- Copilot CLI integration
Cons- Not standalone — requires GitHub account
- No rate limit guarantees
- Limited model selection
- OAuth only (no API key option)
Setup guideDo NOT use `gh auth login`. Run `hermes model` → select GitHub Copilot → follow OAuth device code flow → authenticate in browser → done. Or set `COPILOT_GITHUB_TOKEN` for token-based auth.
🔑 OAuth / Platform
OpenAI Codex
OAuth★★★☆☆
OpenAI Codex CLI integration. OAuth-based access to OpenAI models via Codex subscription. Alternative to direct OpenAI API key billing.
ModelsGPT-4o, GPT-4o-mini (Codex hosted)
PricingIncluded with Codex subscription (~$20/month)
Best forUsers who already subscribe to Codex and want unified billing.
Free tierNo — requires Codex subscription
Pros- Unified billing with Codex
- OAuth — no API key management
- Good model quality
Cons- Requires Codex subscription
- Higher cost than direct API
- Limited model selection
- No free tier
Setup guideRun `hermes login --provider openai-codex` → follow browser auth → models are available via Codex provider
🔑 OAuth / Platform
Nous Portal
OAuth★★★★★
The recommended way to run Hermes Agent. One OAuth covers 300+ frontier models (Claude, GPT, Gemini, DeepSeek, Qwen, Kimi, GLM, MiniMax, Grok) plus the Tool Gateway for web search, image generation, TTS, and browser access.
Models300+ frontier models - Claude, GPT, Gemini, DeepSeek, Qwen, Kimi, GLM, MiniMax, Grok
PricingSubscription-based - see the Nous Portal pricing page
Best forThe documented fast path for new installs (hermes setup --portal). One OAuth for models plus the Tool Gateway.
Free tierNo - subscription OAuth
Pros- 300+ frontier models behind one OAuth
- Includes the Tool Gateway (search, image gen, TTS, browser)
- Documented recommended setup path
- Supports Hermes development
Cons- Subscription required
- OAuth only - no plain API-key path
- Usage subject to portal limits
Setup guideRun `hermes setup --portal` (or `hermes login --provider nous`) → follow the browser OAuth flow → select models via `hermes model`. The same OAuth also unlocks the Tool Gateway.
💻 Local Inference
Ollama (Local)
Local★★★★★
Run LLMs locally on your own hardware. Hundreds of models available — Llama 3.2, Mistral, Phi-4, Gemma, DeepSeek. Zero API cost, fully private.
ModelsHundreds — Llama 3.2, Mistral, Phi-4, Gemma, Qwen, DeepSeek, and community models
Pricing$0 (electricity only)
Best forZero-cost inference, privacy, offline use. 3B models on modest hardware, 7B+ on better machines.
Free tierN/A — it's always free (no API calls)
Pros- Completely free (no API costs)
- Fully private — no data leaves your machine
- Hundreds of models
- Works offline
- Fast for small models on modest hardware
Cons- Requires local hardware (RAM/GPU)
- Small models struggle with complex tasks
- 10-50 tok/s on CPU, 50-100 on GPU
- GGUF file sizes: 2-8GB per model
- Setup needed per machine
Setup guideInstall Ollama: `curl -fsSL https://ollama.com/install.sh | sh` → pull a model: `ollama pull llama3.2` → add a Custom Endpoint with base_url http://localhost:11434/v1 (no key) → select the model in hermes model. For the hosted catalog use the ollama-cloud provider instead.
💻 Local Inference
llama.cpp
Local★★★☆☆
llama.cpp is not a registry provider. It appears in the docs as a local-inference context-size example and as an optional skill; reach its server through the Custom Endpoint flow.
ModelsAny GGUF format model — thousands available on Hugging Face
Pricing$0 (electricity only)
Best forPower users who want full control over inference parameters. Custom quantization levels.
Free tierN/A — always free
Pros- Full control over inference
- Custom quantization
- Runs on CPU efficiently
- Supports any GGUF model
- Lower overhead than Ollama
Cons- More setup than Ollama
- No model download management
- Manual configuration
- Less user-friendly
Setup guideDownload or compile llama.cpp → download GGUF model from Hugging Face → point a Custom Endpoint at the llama.cpp server (for example http://localhost:8080/v1) and select that model
☁️ API Providers
Hugging Face
API★★★☆☆
Hugging Face Inference API and Inference Endpoints. Access to thousands of community and first-party models. Serverless or dedicated endpoints.
ModelsThousands — all models on Hugging Face Hub with Inference API support
PricingServerless: free tier with rate limits. Dedicated endpoints: pay-per-hour for GPU.
Best forAccess to niche or community models not available on other providers. Experimenting with new models.
Free tier✅ Free tier
Config varsHF_TOKEN
Pros- Thousands of models
- Free serverless tier
- Community models available
- Dedicated endpoints for production
Cons- Serverless is slow (cold starts)
- Model quality varies wildly
- Less curated than other providers
- Dedicated endpoints are expensive
Setup guideSign up at huggingface.co → create access token → `echo 'HF_TOKEN="hf_..."' >> ~/.hermes/.env` → select models via `hermes model`
☁️ API Providers
Groq
API★★★☆☆
Groq is no longer in the Hermes LLM provider registry. It survives as an STT/Whisper provider, configured under stt.* with GROQ_API_KEY.
ModelsWhisper speech-to-text models only (no LLM inference)
PricingSTT usage billed by Groq - see console.groq.com
Best forFast, cheap speech-to-text via Whisper. Not usable as a chat model provider.
Free tierCheck Groq STT pricing
Config varsGROQ_API_KEY
Pros- Fast Whisper STT
- Simple API key setup
- Works alongside any LLM provider
Cons- No longer a first-class LLM provider
- STT only - cannot host chat models
- Separate billing from your chat provider
Setup guideSign up at console.groq.com → create API key → `echo 'GROQ_API_KEY="gsk_..."' >> ~/.hermes/.env` → select models via `hermes model`
☁️ API Providers
Together AI
API★★★☆☆
Together AI is not in the Hermes provider registry. Configure it as a custom OpenAI-compatible endpoint using providers: { together: ... } and provider: custom:together.
ModelsLlama 3, Mistral, DeepSeek, Qwen, and other open models
PricingCompetitive with other open-model providers. ~$0.10-1.00/1M tokens.
Best forOpen model inference with fine-tuning capabilities.
Free tier✅ Free tier
Config varsTOGETHER_API_KEY
Pros- Good open model selection
- Fine-tuning API
- Competitive pricing
- Free tier
Cons- Smaller selection than OpenRouter
- No premium/closed models
- Less well-known
Setup guideDefine a custom provider entry in config.yaml with the Together base URL and TOGETHER_API_KEY, then select it with provider: custom:together
☁️ API Providers
Novita AI
API★★★☆☆
Novita AI inference platform. Supports a wide range of open models with competitive pricing. Includes image generation models.
ModelsLlama 3, Mistral, DeepSeek, Stable Diffusion, and others
PricingCompetitive pricing on open models. Image generation additional.
Best forOpen models plus image generation in one provider.
Free tier✅ Free tier
Config varsNOVITA_API_KEY
Pros- Open models + image gen
- Competitive pricing
- Free signup credits
Cons- Newer provider
- Smaller ecosystem
- Variable reliability
Setup guideSign up at novita.ai → create API key → set env var → select models via hermes model
☁️ API Providers
Z.AI / GLM
API★★★☆☆
Zhipu AI's GLM models. Leading Chinese LLM provider. Strong Chinese language performance with competitive English capabilities.
ModelsGLM-4, GLM-4V (vision), GLM-4-Plus
PricingCompetitive with international providers
Best forChinese language tasks. Access to GLM models. Chinese enterprise deployments.
Free tier✅ Free tier
Config varsGLM_API_KEY
Pros- Strong Chinese language
- Vision model available
- Competitive pricing
- Free tier
Cons- Chinese-focused
- English performance lags behind
- Limited international docs
Setup guideSign up at zhipu.ai → create API key → `echo 'GLM_API_KEY="..."' >> ~/.hermes/.env` → select models
☁️ API Providers
MiniMax
API★★★☆☆
MiniMax LLM and TTS models. Chinese provider with competitive language models and high-quality text-to-speech.
ModelsMiniMax models, MiniMax TTS
PricingCompetitive pricing. TTS also available.
Best forChinese language tasks. High-quality Chinese TTS.
Free tier✅ Free tier
Config varsMINIMAX_API_KEY or MINIMAX_CN_API_KEY
Pros- Good Chinese language models
- High-quality TTS
- Competitive pricing
- Free tier
Cons- Chinese-focused
- Smaller model selection
- Limited English support
Setup guideSign up at minimax.com → create API key → set env var → select models
☁️ API Providers
Alibaba / DashScope
API★★★★☆
Alibaba Cloud's Qwen models via DashScope API. Strong Chinese and English performance. Qwen2.5 models are competitive globally.
ModelsQwen2.5-72B, Qwen2.5-Coder, Qwen2-VL (vision), and smaller Qwen models
PricingVery competitive — Qwen models offer excellent value
Best forQwen models — excellent quality-to-price ratio. Both Chinese and English.
Free tier✅ Free tier
Config varsDASHSCOPE_API_KEY
Pros- Qwen models are top-tier
- Excellent value
- Strong bilingual (CN/EN)
- Vision model available
- Coder model for programming
Cons- Alibaba Cloud signup can be complex
- Less community adoption in West
- Documentation mostly in Chinese
Setup guideSign up at dashscope.aliyun.com → create API key → `echo 'DASHSCOPE_API_KEY="sk-..."' >> ~/.hermes/.env` → select models
☁️ API Providers
Kimi / Moonshot
API★★★☆☆
Moonshot AI's Kimi models. Known for very long context windows. Strong Chinese language performance.
ModelsKimi models with long context support
PricingCompetitive
Best forVery long context tasks. Chinese language applications.
Free tier✅ Free tier
Config varsKIMI_API_KEY
Pros- Very long context windows
- Good Chinese performance
- Competitive pricing
- Free tier
Cons- Chinese-focused
- Smaller international presence
- Limited model variety
Setup guideSign up at moonshot.cn → create API key → set env var → select models
☁️ API Providers
Fireworks AI
API★★★★☆
Fireworks AI - fast inference for open and proprietary models, using native slash-form catalog model IDs.
Modelsaccounts/fireworks/models/kimi-k2p6 and the live Fireworks catalog
PricingPay-per-token - see fireworks.ai pricing
Best forFast inference for open models, with a managed Nous fallback path.
Free tierNot verified against a fetched source - see the provider docs
Config varsF, I, R, E, W, O, R, K, S, _, A, P, I, _, K, E, Y
Pros- Fast inference
- Native catalog model IDs
- First-class provider ID with aliases fw / fireworks-ai
Cons- Open-model catalog only
- Base URL changes go through model.base_url in config.yaml, not .env
Setup guideCreate an API key at fireworks.ai → `echo 'FIREWORKS_API_KEY="..."' >> ~/.hermes/.env` → `hermes chat --provider fireworks`
☁️ API Providers
Arcee AI
API★★★☆☆
Arcee AI - the Trinity model family.
Modelstrinity-large-thinking
PricingPay-per-token - see arcee.ai pricing
Best forTrinity models from Arcee AI.
Free tierNot verified against a fetched source - see the provider docs
Config varsA, R, C, E, E, A, I, _, A, P, I, _, K, E, Y
Pros- Trinity model family
- First-class provider ID with aliases arcee-ai / arceeai
Cons- Smaller catalog than aggregators
Setup guideCreate an API key at arcee.ai → add ARCEEAI_API_KEY to ~/.hermes/.env → `hermes chat --provider arcee`
☁️ API Providers
GMI Cloud
API★★★★☆
GMI Cloud - open and reasoning models via an OpenAI-compatible API.
Modelsdeepseek-ai/DeepSeek-V3.2 and other catalog IDs from GMI /v1/models
PricingPay-per-token - see gmicloud.ai pricing
Best forOpen and reasoning models from GMI Cloud with an OpenAI-compatible endpoint.
Free tierNot verified against a fetched source - see the provider docs
Config varsG, M, I, _, A, P, I, _, K, E, Y
Pros- OpenAI-compatible API
- Reasoning models available
- Override base URL with GMI_BASE_URL
Cons- Model IDs must come from the GMI /v1/models endpoint
Setup guideCreate an API key at gmicloud.ai → add GMI_API_KEY to ~/.hermes/.env → `hermes chat --provider gmi --model deepseek-ai/DeepSeek-V3.2`
☁️ API Providers
Xiaomi MiMo
API★★★☆☆
Xiaomi MiMo models via the Xiaomi API.
Modelsmimo-v2-pro
PricingPay-per-token - see Xiaomi API pricing
Best forMiMo models from Xiaomi.
Free tierNot verified against a fetched source - see the provider docs
Config varsX, I, A, O, M, I, _, A, P, I, _, K, E, Y
Pros- First-class provider ID with aliases mimo / xiaomi-mimo
- Override base URL with XIAOMI_BASE_URL
Cons- Newer provider with a smaller catalog
Setup guideObtain a Xiaomi API key → add XIAOMI_API_KEY to ~/.hermes/.env → `hermes chat --provider xiaomi --model mimo-v2-pro`
☁️ API Providers
Tencent TokenHub
API★★★☆☆
Tencent TokenHub - Tencent-hosted models (Hy3 Preview and others).
Modelshy3-preview
PricingPay-per-token - see Tencent Cloud pricing
Best forTencent-hosted models, including Hy3 Preview.
Free tierNot verified against a fetched source - see the provider docs
Config varsT, O, K, E, N, H, U, B, _, A, P, I, _, K, E, Y
Pros- First-class provider ID with aliases tencent / tokenhub / tencentmaas
- Override base URL with TOKENHUB_BASE_URL
Cons- China-focused catalog
- Documentation largely Chinese
Setup guideObtain a TokenHub API key → add TOKENHUB_API_KEY to ~/.hermes/.env → `hermes chat --provider tencent-tokenhub --model hy3-preview`
☁️ API Providers
Kilo Code
API★★★☆☆
Kilo Code provider.
ModelsKilo Code catalog models
PricingNot verified against a fetched source - see the provider pricing page
Best forCoding-focused model access through Kilo Code.
Free tierNot verified against a fetched source - see the provider docs
Config varsK, I, L, O, C, O, D, E, _, A, P, I, _, K, E, Y
Pros- Coding-oriented catalog
- Simple API key setup
Cons- Niche provider with a smaller catalog
Setup guideCreate an API key with Kilo Code → add KILOCODE_API_KEY to ~/.hermes/.env → `hermes chat --provider kilocode`
☁️ API Providers
StepFun
API★★★☆☆
StepFun Step-series models via an OpenAI-compatible API.
Modelsstep-3.5-flash
PricingPay-per-token - see platform.stepfun.com pricing
Best forStep-series models from StepFun.
Free tierNot verified against a fetched source - see the provider docs
Config varsS, T, E, P, F, U, N, _, A, P, I, _, K, E, Y
Pros- OpenAI-compatible API
- Override base URL with STEPFUN_BASE_URL
- Also registered as a Step Plan provider plugin
Setup guideCreate an API key at platform.stepfun.com → add STEPFUN_API_KEY to ~/.hermes/.env → `hermes chat --provider stepfun --model step-3.5-flash`
☁️ API Providers
Upstage Solar
API★★★☆☆
Upstage Solar API provider.
ModelsUpstage Solar models
PricingNot verified against a fetched source - see the provider pricing page
Best forSolar models from Upstage.
Free tierNot verified against a fetched source - see the provider docs
Config varsU, P, S, T, A, G, E, _, A, P, I, _, K, E, Y
Pros- First-class provider plugin
- Solar model family
Cons- Smaller ecosystem outside Korea
Setup guideCreate an Upstage API key → add UPSTAGE_API_KEY to ~/.hermes/.env → `hermes chat --provider upstage`
☁️ API Providers
NVIDIA NIM
API★★★★☆
NVIDIA Build / NIM - Nemotron and other open models hosted on build.nvidia.com, or a local NIM endpoint.
Modelsnvidia/nemotron-3-super-120b-a12b
PricingFree API key at build.nvidia.com; billed by NVIDIA beyond that
Best forOpen models on NVIDIA infrastructure, with an optional local NIM endpoint for on-prem GPUs.
Free tier✅ Free tier
Config varsN, V, I, D, I, A, _, A, P, I, _, K, E, Y
Pros- Free API key from build.nvidia.com
- Local NIM endpoint supported via NVIDIA_BASE_URL
- Billing-origin header attached automatically
Cons- Catalog skews to NVIDIA-hosted open models
- Local NIM needs your own GPU
Setup guideGet a free API key at build.nvidia.com → add NVIDIA_API_KEY to ~/.hermes/.env → `hermes chat --provider nvidia`. For local NIM set NVIDIA_BASE_URL=http://localhost:8000/v1
☁️ API Providers
AWS Bedrock
API★★★★☆
AWS Bedrock - Claude, Amazon Nova, DeepSeek v3.2, Meta Llama 4 and other models through the Converse API. Uses the standard boto3 credential chain instead of a Hermes API key.
Modelsus.anthropic.claude-sonnet-4-6 and other Bedrock model IDs
PricingBilled through AWS (Bedrock pricing)
Best forTeams already on AWS who want Claude, Nova, DeepSeek, and Llama behind one credential chain.
Free tierNo - AWS billing required
Config varsA, W, S, _, P, R, O, F, I, L, E
Pros- No separate API key - reuses AWS credentials
- Converse API translates across Claude, Nova, DeepSeek, and Llama
- IAM roles work on EC2, ECS, and Lambda
Cons- Requires an AWS account and Bedrock model access
- Region and cross-region inference setup add complexity
Setup guideConfigure AWS credentials (profile, env vars, IAM role, or SSO) → `hermes chat --provider bedrock --model us.anthropic.claude-sonnet-4-6`, or set bedrock.region in config.yaml
☁️ API Providers
Google Vertex AI
API★★★★☆
Google Vertex AI - Gemini models through the Vertex OpenAI-compatible endpoint. Authentication is OAuth2: short-lived access tokens minted from a service-account JSON or Application Default Credentials, with no static API key.
Modelsgoogle/gemini-3-flash-preview (Vertex requires the google/ prefix)
PricingBilled through GCP (Vertex AI pricing)
Best forGCP-based deployments that want Gemini behind service-account auth and GCP billing.
Free tierNo - GCP billing required
Config varsV, E, R, T, E, X, _, C, R, E, D, E, N, T, I, A, L, S, _, P, A, T, H
Pros- No static API key - short-lived OAuth2 tokens auto-refresh
- Service-account or ADC authentication
- Project and region live in config.yaml
Cons- Requires a GCP project with billing
- Gemini 3.x previews require the global region
- Hermes lazy-installs google-auth on first use
Setup guideSet VERTEX_CREDENTIALS_PATH to a service-account JSON, or run `gcloud auth application-default login` → `hermes model` and pick Google Vertex AI → choose project, region, and model
☁️ API Providers
Azure AI Foundry
API★★★☆☆
Azure AI Foundry provider, using an Azure OpenAI / Foundry endpoint and key.
ModelsAzure OpenAI / Foundry deployed models
PricingBilled through Azure - see Azure OpenAI pricing
Best forOrganizations standardized on Azure that need deployments inside their tenant.
Free tierNot verified against a fetched source - see the provider docs
Config varsA, z, u, r, e, , O, p, e, n, A, I, , e, n, d, p, o, i, n, t, , +, , k, e, y, , (, e, n, t, e, r, e, d, , i, n, , h, e, r, m, e, s, , m, o, d, e, l, ;, , n, o, , n, a, m, e, d, , e, n, v, , v, a, r, , i, n, , t, h, e, , d, o, c, s, )
Pros- Runs in your own Azure tenant
- Enterprise compliance and networking options
- Works with Azure OpenAI deployments
Cons- Requires an Azure subscription and model deployment
- Endpoint and key management is per-deployment
Setup guideDeploy a model in Azure AI Foundry → `hermes model` and pick Azure AI Foundry → supply the endpoint and key
☁️ API Providers
Ollama Cloud
API★★★★☆
Ollama Cloud hosts the same open-weight catalog as local Ollama without the GPU requirement. The model catalog is fetched dynamically from ollama.com/v1/models and cached for one hour.
Modelsgpt-oss:120b, glm-4.6:cloud, qwen3-coder:480b-cloud and the rest of the Ollama catalog
PricingSubscription or API usage billed by Ollama
Best forLarge open-weight models you cannot run locally, without managing a GPU host.
Free tierNot verified against a fetched source - see the provider docs
Config varsO, L, L, A, M, A, _, A, P, I, _, K, E, Y
Pros- No GPU required
- Same open-weight catalog as local Ollama
- Live catalog discovery
- model:tag notation preserved
Cons- Cloud dependency and billing
- Requires network access
- Different provider ID from local Ollama (use the Custom Endpoint flow for local)
Setup guidePick Ollama Cloud in `hermes model` → paste your API key from ollama.com/settings/keys → select from the discovered models. Or set model.provider to ollama-cloud in config.yaml
🔑 OAuth / Platform
Qwen OAuth (Portal)
OAuth★★★★☆
Alibaba Qwen Portal with browser-based OAuth login. Sign in through the browser and Hermes persists the refresh token in ~/.hermes/auth.json.
Modelsqwen3-coder-plus and other Qwen Portal models
PricingConsumer Qwen Portal account - see Alibaba portal terms
Best forIndividual users who want Qwen models without managing an API key.
Free tierOAuth via an Alibaba account
Config varsN, o, n, e, , (, b, r, o, w, s, e, r, , P, K, C, E, , l, o, g, i, n, )
Pros- No API key to manage
- Refresh token persisted automatically
- Consumer-facing Qwen Portal models
Cons- OAuth only - no API-key path
- Set HERMES_QWEN_BASE_URL only if the portal endpoint relocates
- Different from the alibaba DashScope provider
Setup guidePick Qwen OAuth (Portal) in `hermes model` → complete the browser sign-in → confirm; credentials are saved to ~/.hermes/auth.json. The endpoint defaults to portal.qwen.ai/v1
🔑 OAuth / Platform
MiniMax (OAuth)
OAuth★★★★☆
MiniMax OAuth - browser login for MiniMax-M2.7 and MiniMax-M2.7-highspeed, with no API key and no billing setup required. Uses the Anthropic Messages-compatible endpoint under the hood.
ModelsMiniMax-M2.7 (main), MiniMax-M2.7-highspeed (default auxiliary model)
PricingNo billing setup required - consumer OAuth login
Best forMiniMax access for individuals without an API key or billing setup.
Free tier✅ Free tier
Config varsN, o, n, e, , (, b, r, o, w, s, e, r, , P, K, C, E, , l, o, g, i, n, )
Pros- No API key or billing setup
- OAuth tokens persisted automatically
- M2.7-highspeed wired as the default auxiliary model
Cons- OAuth only - ignores MINIMAX_API_KEY and MINIMAX_BASE_URL
- Limited model list
Setup guidePick MiniMax (OAuth) in `hermes model` → sign in through the browser (global or CN region) → confirm; credentials are saved to ~/.hermes/auth.json. Requests go to api.minimax.io/anthropic
🔑 OAuth / Platform
xAI Grok OAuth
OAuth★★★★☆
xAI Grok OAuth for SuperGrok and X Premium+ subscribers - browser login instead of an API key. The same bearer token is reused by the direct-to-xAI tools (TTS, image gen, video gen, transcription).
ModelsCurrent Grok models (see x.ai)
PricingIncluded with a SuperGrok or X Premium+ subscription
Best forSubscribers who want Grok without paying for API usage separately.
Free tier✅ Free tier
Config varsN, o, n, e, , (, b, r, o, w, s, e, r, , O, A, u, t, h, )
Pros- No API key
- Token reused by TTS, image gen, video gen, and transcription
- SuperGrok and X Premium+ supported
Cons- Requires a subscription
- Remote hosts need an ssh port-forward for the OAuth callback
Setup guidePick xAI Grok OAuth (SuperGrok / Premium+) in `hermes model`, or run `hermes auth add xai-oauth` → complete the browser flow. On a remote host set up an ssh -L tunnel for the OAuth callback
☁️ API Providers
MiniMax (China)
API★★★☆☆
MiniMax against the China endpoint, using a separate API key from the global endpoint.
ModelsMiniMax-M2.7
PricingPay-per-token - see MiniMax pricing
Best forMainland-China deployments that must stay on the CN endpoint.
Free tierNot verified against a fetched source - see the provider docs
Config varsM, I, N, I, M, A, X, _, C, N, _, A, P, I, _, K, E, Y
Pros- Endpoint close to CN users
- Independent key from the global endpoint
- Override base URL with MINIMAX_CN_BASE_URL
Cons- CN billing and console only
- Separate key management from the global provider
Setup guideCreate a key in the MiniMax China console → add MINIMAX_CN_API_KEY to ~/.hermes/.env → `hermes chat --provider minimax-cn --model MiniMax-M2.7`
☁️ API Providers
Kimi / Moonshot (China)
API★★★☆☆
Moonshot Kimi provider against the China endpoint api.moonshot.cn, distinct from the international api.moonshot.ai endpoint.
Modelskimi-k2.5
PricingPay-per-token - see Moonshot pricing
Best forCN-region Kimi access with a separate key.
Free tierNot verified against a fetched source - see the provider docs
Config varsK, I, M, I, _, C, N, _, A, P, I, _, K, E, Y
Pros- CN endpoint with lower latency for regional users
- First-class provider ID with aliases kimi-cn / moonshot-cn
Cons- Separate key from the international kimi-coding provider
Setup guideCreate a key at platform.moonshot.cn → add KIMI_CN_API_KEY to ~/.hermes/.env → `hermes chat --provider kimi-coding-cn --model kimi-k2.5`
☁️ API Providers
Alibaba Cloud (Coding Plan)
API★★★★☆
Alibaba Cloud Coding Plan - a pricing SKU separate from standard DashScope API access, exposed as its own first-class provider with a different base URL and billing surface.
Modelsqwen3-coder-plus
PricingCoding Plan subscription SKU - see Alibaba Cloud pricing
Best forSubscribers to Alibaba Coding Plan who want Qwen coder models billed through that SKU.
Free tierDepends on the Coding Plan subscription
Config varsD, A, S, H, S, C, O, P, E, _, A, P, I, _, K, E, Y
Pros- Reuses the existing DASHSCOPE_API_KEY
- Separate billing SKU with its own endpoint
- Fixes silent fall-through to OpenRouter when the provider was unregistered
Cons- Requires a Coding Plan subscription
- Endpoint differs from standard DashScope
Setup guideSubscribe to the Alibaba Coding Plan → reuse your existing DASHSCOPE_API_KEY → set model.provider to alibaba_coding (alias for alibaba-coding-plan) with model qwen3-coder-plus
☁️ API Providers
OpenCode Zen
API★★★☆☆
OpenCode Zen provider.
ModelsOpenCode Zen catalog models
PricingNot verified against a fetched source - see the provider pricing page
Best forModel access through the OpenCode Zen gateway.
Free tierNot verified against a fetched source - see the provider docs
Config varsO, P, E, N, C, O, D, E, _, Z, E, N, _, A, P, I, _, K, E, Y
Pros- First-class provider plugin
- Integrates with the OpenCode ecosystem
Cons- Smaller catalog than aggregators
Setup guideCreate an OpenCode Zen key → add OPENCODE_ZEN_API_KEY to ~/.hermes/.env → `hermes chat --provider opencode-zen`
☁️ API Providers
OpenCode Go
API★★★☆☆
OpenCode Go provider.
ModelsOpenCode Go catalog models
PricingNot verified against a fetched source - see the provider pricing page
Best forModel access through the OpenCode Go gateway.
Free tierNot verified against a fetched source - see the provider docs
Config varsO, P, E, N, C, O, D, E, _, G, O, _, A, P, I, _, K, E, Y
Pros- First-class provider plugin
- Separate key from OpenCode Zen
Cons- Smaller catalog than aggregators
Setup guideCreate an OpenCode Go key → add OPENCODE_GO_API_KEY to ~/.hermes/.env → `hermes chat --provider opencode-go`
💻 Local Inference
LM Studio
Local★★★★☆
LM Studio local server - run any GGUF model locally and expose it over an OpenAI-compatible API at http://127.0.0.1:1234/v1.
ModelsAny GGUF model served by LM Studio
Pricing$0 (electricity only)
Best forDesktop users who prefer a GUI model manager over the Ollama CLI.
Free tierN/A - local inference
Config varsL, M, _, A, P, I, _, K, E, Y
Pros- GUI model management
- Full control over context and sampling
- Reachable through the Custom Endpoint flow too
Cons- Local hardware limits model size
- Requires the LM Studio server to stay running
- Optional LM_API_KEY needed if the server requires auth
Setup guideInstall LM Studio and load a model → enable the local server → `hermes model` and pick LM Studio. Optional LM_API_KEY; override the URL with LM_BASE_URL
💻 Local Inference
Custom Endpoint
Local★★★★★
Custom / self-hosted OpenAI-compatible endpoint - point Hermes at any server implementing /v1/chat/completions. This is also the documented path for local Ollama (base_url http://localhost:11434/v1, no key).
ModelsAny model your endpoint serves
Pricing$0 for self-hosted hardware; endpoint-specific for hosted routers
Best forLocal models via vLLM, llama.cpp, Ollama, LM Studio, or any third-party OpenAI-compatible API.
Free tierNot verified against a fetched source - see the provider docs
Config varsm, o, d, e, l, ., b, a, s, e, _, u, r, l
Pros- Works with any OpenAI-compatible server
- No plugin or provider registration needed
- The documented path for local Ollama and llama.cpp
Cons- You manage the server, model files, and context limits
- Not a registry provider, so no catalog discovery
Setup guidePick Custom endpoint in `hermes model`, then enter the API base URL, optional API key, and model name; it is saved under the providers section of config.yaml
🔑 OAuth / Platform
GitHub Copilot (ACP)
OAuth★★★☆☆
GitHub Copilot via ACP - spawns a local copilot --acp --stdio subprocess instead of calling the Copilot API directly.
ModelsCopilot-served models available to the local CLI
PricingIncluded with the GitHub Copilot subscription
Best forUsers whose Copilot access must run through the local CLI rather than the API.
Free tier✅ Free tier
Config varsN, o, n, e, , (, u, s, e, s, , t, h, e, , l, o, c, a, l, , c, o, p, i, l, o, t, , C, L, I, , s, e, s, s, i, o, n, )
Pros- Reuses an existing Copilot CLI session
- No API token handling in Hermes
Cons- Requires the copilot binary in PATH
- Subprocess transport is slower to start than a direct API call
Setup guideInstall the GitHub Copilot CLI and run `copilot login` → pick GitHub Copilot (ACP) in `hermes model`. Override the binary with HERMES_COPILOT_ACP_COMMAND and args with HERMES_COPILOT_ACP_ARGS
☁️ API Providers
Ramp Router
API★★★☆☆
Ramp Router provider.
ModelsModels routed by Ramp Router
PricingNot verified against a fetched source - see the provider pricing page
Best forRouting requests across models through Ramp Router.
Free tierNot verified against a fetched source - see the provider docs
Config varsR, A, M, P, _, R, O, U, T, E, R, _, A, P, I, _, K, E, Y
Pros- First-class provider plugin
- Router picks the backing model
Cons- Routing behaviour is defined by the service, not Hermes
Setup guideCreate a Ramp Router key → add RAMP_ROUTER_API_KEY to ~/.hermes/.env → `hermes chat --provider router`
☁️ API Providers
DeepInfra
API★★★★☆
DeepInfra - 100+ open models, pay-per-use, behind an OpenAI-compatible API.
Models100+ open models from the DeepInfra catalog
PricingPay-per-use - see deepinfra.com pricing
Best forBroad open-model access with per-use billing.
Free tierNot verified against a fetched source - see the provider docs
Config varsD, E, E, P, I, N, F, R, A, _, A, P, I, _, K, E, Y
Pros- Large open-model catalog
- Pay-per-use with no subscription
- Also powers the DeepInfra image and video backends
Cons- Open models only
- Catalog quality varies by model
Setup guideCreate a DeepInfra API key → add DEEPINFRA_API_KEY to ~/.hermes/.env → `hermes chat --provider deepinfra`