Hermes Agent
Connect Hermes Agent to PromptsForLess as a custom endpoint, add a model alias and set main and auxiliary models.
Hermes Agent works with custom OpenAI-compatible endpoints. For agent work, pick a model with tool calling and enough context for a full session. Hermes' quickstart asks for at least 64,000 context tokens.
Add a custom endpoint
Install Hermes with its official instructions, then run:
hermes modelChoose Custom Endpoint and enter:
| Field | Value |
|---|---|
| Base URL | https://api.promptsforless.com/v1 |
| API key | Your PromptsForLess key from the dashboard |
| Model | A model ID from the catalog, such as deepseek-v4.1-flash |
The picker writes the right configuration for your installed version, so prefer it over hand-editing.
Add a model alias
Hermes also supports custom endpoint aliases. With PFL_API_KEY available to the Hermes process, merge this alias into your configuration:
model_aliases:
pfl:
model: deepseek-v4.1-flash
provider: custom
base_url: https://api.promptsforless.com/v1
key_env: PFL_API_KEYRun /model pfl inside a chat to switch to it. Hermes' model configuration guide documents this format.
Check main and auxiliary models
Run hermes status to see your provider and model. Send a short prompt, then ask Hermes to read a harmless file in a test folder. Review each action before you approve it.
Auxiliary slots have their own models. Vision needs a model that accepts images. Context compression needs a model with enough context for the history it summarizes. A new default applies to new sessions; switch existing chats with /model.
Hermes keeps settings in config.yaml and credentials in ~/.hermes/.env. Its configuration reference explains credential storage and environment substitution. Never commit either file.
Troubleshooting
| Symptom | Fix |
|---|---|
| Custom provider missing | Make the key available to Hermes and reopen the model picker. |
| Context rejected | Pick a model that meets Hermes' minimum and set its real context limit. |
| A chat still uses the old model | Start a new session or run /model in the current one. |
| Auxiliary task fails | Check that slot's endpoint, model and capabilities on their own. |
See the model catalog for prices and the quickstart for direct request checks.