Hermes Agent: connect to ModelFold

Higher credit usage on the first session is normal

Agents attach fixed tool definitions and a system prompt to every request, often tens of thousands of tokens. On the first session this content is written to the cache and billed as full input, so credit usage is higher. Later requests hit the cache and are billed at the much lower cached rate, so usage drops noticeably. This is expected behavior — you are not being overcharged. For high-volume usage, please contact support — we can offer discounts.

All coding tools · More integrations · Unsupported and limited tools

Configured from official documentation; not tested in this client. This verifies configuration availability, not every agent feature or model.

Install and prepare

Install the official release using the installation link in the vendor documentation. For an editor extension, use the official publisher in your editor marketplace. Create a key in ModelFold API Keys; keep it out of project files and screenshots. Where a snippet references MODELFOLD_API_KEY, set it in the environment before launching the tool:

export MODELFOLD_API_KEY="YOUR_MODELFOLD_API_KEY"

Configure the provider

Setup location: hermes model / ~/.hermes/config.yaml. Verification: Configured from official documentation; not tested in this client.

Run hermes model and select a custom endpoint. Configure the main model, not just an auxiliary model. Select Anthropic Messages for Claude.

providers:
  modelfold:
    api: https://modelfold.ai/anthropic
    key_env: MODELFOLD_API_KEY
    transport: anthropic_messages
    default_model: claude-opus-5-5
model:
  provider: modelfold
  default: claude-opus-5-5

For Claude, prefer the native Anthropic route. Cache savings require eligible repeated prefixes and client cache support; a custom endpoint alone does not guarantee a cache hit.

Official configuration reference.

Model and endpoint

Field	Value
Endpoint used by this guide	https://modelfold.ai/anthropic
API model ID	claude-opus-5-5
Credential	Your ModelFold API key

The model ID must match All model IDs. Client prefixes such as openai/ select an adapter and are not part of the server model ID. Do not add /v1 or /messages twice; follow this tool's example.

Verify

Start a new chat, select the configured provider/model, and send: Reply with exactly OK. Do not use tools. A successful response verifies that model route only. Then check the corresponding request in ModelFold usage. A model appearing in the picker alone is not a connectivity test. Before a long coding task, try one small change in a disposable project.

Troubleshooting

Symptom	Check
401	Use a ModelFold key, not a vendor subscription token; remove extra whitespace.
404	Check the protocol, URL suffix and exact model ID.
Empty model picker	Enter the documented ID manually when supported; restart the client after changing environment variables.
Chat works, tools fail	Select a tool-capable model; client agent features are not guaranteed by basic chat compatibility.
Requests still use the built-in service	Select your custom provider explicitly and check saved settings/environment overrides.

Prices follow the model price page. Tasks can make multiple requests. Final deductions are shown in your account; client cost estimates are not the billing record.