Hermes Agent connects to Dahl as a custom OpenAI-compatible endpoint. It needs the base URL with /v1, a key with tokens on it, and a model ID.
Setup
- Get a key with tokens — in /account → API keys, Allocate tokens from the pool to a key (a new key starts at 0).
- Run the model wizard in a terminal (not inside a Hermes chat) and choose Custom endpoint:
hermes model # → Custom endpoint # Base URL: https://inference.dahl.global/v1 # API key: dahl_… # Model: MiniMaxAI/MiniMax-M2.7
If the wizard asks for the API mode, choose Chat Completions. Set the context length to the model window (see below).
Config file
The wizard saves the settings to ~/.hermes/config.yaml. You can also edit it by hand and keep the key in ~/.hermes/.env:
# ~/.hermes/config.yaml model: default: MiniMaxAI/MiniMax-M2.7 provider: custom base_url: https://inference.dahl.global/v1 key_env: DAHL_API_KEY context_length: 180000
# ~/.hermes/.env DAHL_API_KEY=dahl_…
Context length
Hermes needs at least 64K tokens of context for agent work with tools. Dahl models fit: MiniMax M2.7 — 180K, DeepSeek V4 Flash and GLM 5.3 Flash — 400K. Start with MiniMaxAI/MiniMax-M2.7; current IDs are in GET /v1/models.
Common errors
| Message | Cause | Fix |
|---|---|---|
available tokens exhausted (402) | The key has 0 tokens | Allocate from the pool in /account |
Method Not Allowed (405) | Base URL without /v1 | Use https://inference.dahl.global/v1 |
This model is not currently offered (400) | Wrong or stale model ID | Use an exact id from GET /v1/models |
This model is at concurrency capacity (429) | The model is full at peak hours | Retry after Retry-After or switch model. See Errors |