Agent Harness
Connect Hermes to SoCLaaS
Point a Hermes agent harness at SoCLaaS so its planning and multi-step tool use run on your gateway key and model.
No generated config for Hermes. The portal's Client Setup page provides ready-made configs for Codex, OpenCode, and Qwen Code. For Hermes, use the manual connection values below.
Hermes configurations differ by version and deployment. Use the connection values below in your version's OpenAI-compatible provider settings, then verify against the Hermes documentation.
Connection values
| Setting | Value |
|---|---|
| Provider type | OpenAI-compatible |
| Base URL | https://soclaas-api.comp.nus.edu.sg/v1 |
| API key | From the SoCLaaS portal, stored in SOCLAAS_API_KEY |
| Model | A chat-capable ID from GET /v1/models |
Steps
-
Get an API key
Sign in to the SoCLaaS portal from the NUS network and create a personal key.
-
Export the key
export SOCLAAS_API_KEY="YOUR_API_KEY" -
Find a model ID
curl https://soclaas-api.comp.nus.edu.sg/v1/models \ -H "Authorization: Bearer $SOCLAAS_API_KEY"Choose a chat-capable model. If Hermes relies on tool calling, confirm the model supports it before running long tasks.
-
Configure the provider
Open Hermes' model or provider configuration and select an OpenAI-compatible provider. Enter the base URL, key reference, and model ID from the table.
Keep the key in
SOCLAAS_API_KEYif Hermes supports environment-based secrets. Otherwise store it in the local config and exclude that file from version control. -
Run a smoke test
Ask Hermes for a short response with no tools first, then run a task that calls one tool. That separates connection problems from tool-support problems.
Troubleshooting
- 401 or 403: confirm the harness can read
SOCLAAS_API_KEY, or that the inline key is current. - Tool calls fail: switch to a model with tool-calling support.
- Timeout: check your network connection and the gateway status.
- 404 on requests: check that the base URL ends in
/v1and does not repeat the path.