A Hermes Agent provider plugin for CrofAI — powerful open-source LLMs at crazy cheap pricing.
Adds CrofAI as a first-class provider in Hermes, with auto-detected model listings, credential management, and full integration with hermes doctor, hermes model, and hermes setup.
curl -fsSL https://raw.githubusercontent.com/FloorIsGround/hermes-crofai/main/install.sh | bashDrop these two files into your Hermes plugins directory:
mkdir -p ~/.hermes/plugins/model-providers/crofai
curl -fsSL https://raw.githubusercontent.com/FloorIsGround/hermes-crofai/main/plugin.yaml \
-o ~/.hermes/plugins/model-providers/crofai/plugin.yaml
curl -fsSL https://raw.githubusercontent.com/FloorIsGround/hermes-crofai/main/__init__.py \
-o ~/.hermes/plugins/model-providers/crofai/__init__.py- Get an API key — sign up at crof.ai/signin
- Add your key to
~/.hermes/.env:CROFAI_API_KEY="your-key-here" - Restart Hermes and select CrofAI:
Or set it as your default:
hermes --provider crofai
hermes config set model.provider crofai hermes config set model.default deepseek-v4-flash
Run hermes doctor — you should see CrofAI listed under API Connectivity with a check mark.
The model list is fetched live from CrofAI's API on provider selection. Current models include:
| Model | Context | Quant |
|---|---|---|
| DeepSeek V4 Pro | 1M | Q4_0 |
| DeepSeek V4 Pro (Precision) | 1M | Q8_0 |
| DeepSeek V4 Flash | 1M | Q4_0 |
| DeepSeek V3.2 | 164K | Q4_0 |
| Kimi K2.6 | 262K | Q3_K_L |
| Kimi K2.6 (Precision) | 262K | int4 |
| Kimi K2.5 | 262K | Q4_K_M |
| Kimi K2.5 (Lightning) | 131K | 530b-int4 |
| GLM 5.1 | 203K | Q6_K |
| GLM 5.1 (Precision) | 203K | Q8_0 |
| GLM 5 | 203K | Q4_0 |
| GLM 4.7 | 203K | Q8_0 |
| GLM 4.7 Flash | 203K | FP8 |
| Gemma 4 31B | 262K | Q4_0 |
| MiniMax M2.5 | 205K | AWQ |
| Qwen3.6 27B | 262K | Q4_0 |
| Qwen3.5 397B A17B | 262K | Q4_0 |
| Qwen3.5 9B | 262K | FP8 |
| MiMo V2.5 Pro | 1M | Q4_0 |
| MiMo V2.5 Pro (Precision) | 1M | Q8_0 |
Run hermes model after selecting CrofAI to see the full live list with pricing.
- Chat completions (streaming and non-streaming)
- Tool/function calling
- Structured outputs (JSON schema)
- Vision (on supported models)
- Reasoning effort control (
low,medium,high,none) - All standard parameters (
temperature,top_p,max_tokens,stop,seed,repetition_penalty)
hermes-crofai/
├── __init__.py # Provider profile definition
├── plugin.yaml # Plugin manifest
├── install.sh # One-liner install script
└── README.md # This file
MIT