How to use Hermes Agent with LLMBase

Hermes Agent can use a custom OpenAI-compatible endpoint for its main model. LLMBase exposes a dedicated Agent API for this purpose, allowing Hermes to use the current LLMBase agent model catalog with a separately revocable chat-agent key.

This is an agent integration, not a direct inference API setup. Use a Pro or Business Chat/Agent subscription and a key beginning with llmbase_chat_.... If an EU-hosted model is required, select an eligible model marked EU-hosted in the LLMBase model catalog. The model choice matters because processing location depends on the specific model.

1. Create a LLMBase chat-agent key

Open Dashboard → Agent Access in LLMBase and create a new chat-agent key. Copy it when it is shown and keep it private.

Setting	Value
Key prefix	llmbase_chat_...
Agent API base URL	https://llmbase.ai/api/v1/agents
Models endpoint	GET /models
Chat endpoint	POST /chat/completions

Do not substitute an llmbase_... inference key. That key works with the direct Inference API at https://api.llmbase.ai/v1; it is not valid for the Agent API in this guide.

2. Pick a current model that is suitable for Hermes

Hermes is designed for multi-step work, so model selection should be deliberate. Retrieve the current agent-capable catalog first:

curl https://llmbase.ai/api/v1/agents/models \
  -H "Authorization: Bearer $LLMBASE_CHAT_AGENT_KEY"

Choose an ID from that response. Hermes currently requires a model with at least 64,000 tokens of context for its longer agent workflows, so verify context length before configuring it. For coding and tool-based work, also check the model’s available tool support and output limit.

If the task has an EU-hosting requirement, cross-check the returned model against the LLMBase model catalog and select an eligible model that is currently labelled EU-hosted.

3. Configure Hermes with the LLMBase agent endpoint

Hermes supports a custom endpoint through its model configuration. Keep the secret in Hermes’ user-level .env file and refer to it from config.yaml rather than embedding it in a project repository.

First, put the agent key in ~/.hermes/.env:

LLMBASE_CHAT_AGENT_KEY=llmbase_chat_...

Then configure the main model in ~/.hermes/config.yaml:

model:
  default: <model-id-from-agent-models>
  provider: custom
  base_url: https://llmbase.ai/api/v1/agents
  api_key: ${LLMBASE_CHAT_AGENT_KEY}

Replace <model-id-from-agent-models> with an exact ID returned by the LLMBase Agent API. Hermes expands ${LLMBASE_CHAT_AGENT_KEY} from its user-level environment configuration, so the secret stays separate from the regular config.

You can also run hermes model and choose a custom endpoint interactively. For a repeatable setup, compare the values it saves with the table above: the key must be an llmbase_chat_... key, and the base URL must be the LLMBase Agent API.

4. Verify the agent before giving it real authority

Start Hermes in a non-sensitive project and ask for a read-only result:

Inspect this repository. Explain its test setup, identify the most likely entry point, and propose the next three checks. Do not edit files or run commands.

Check that Hermes uses the intended model and that the response is useful. Then build up the tool scope in small steps. A productive coding agent should first understand a project, then suggest a change, then run a clearly reviewed test. It should not start with broad file access, production credentials, or actions that change external systems.

Good operational controls include:

• Keep ~/.hermes/.env private and out of backups or repositories that other people can read.
• Use one LLMBase agent key per person or environment when you need independent revocation.
• Keep the agent’s working directory limited to the project you intend it to inspect.
• Review generated patches and commands before accepting them.
• Revoke the key immediately if it may have been exposed.

5. Use features only when the chosen model supports them

Hermes can use a model for complex workflows, but the API request still has to match the model’s declared capabilities. Before enabling a particular workflow, check whether the selected LLMBase model supports:

• Tool calls for agent tool loops.
• A suitable context window for the task.
• Structured output for machine-readable handoffs.
• Image input if the agent will inspect screenshots or diagrams.
• A supported reasoning setting if your Hermes configuration sends one.

If a setting is rejected, do not try to force it through. Select a model whose published capabilities match the task or simplify the request to the documented compatibility surface.

EU-hosted model selection for Hermes

For a project that requires an EU-hosted model, make the selection explicit in config.yaml and in the team’s project notes. That makes it easier to review the setup and to notice when a model needs to be replaced.

Use current public LLMBase documentation for privacy, data-residency, and contractual decisions. An EU-hosted model label is a useful part of a controlled setup, but it does not remove the need to assess the specific data, workflow, and organisation policy involved.

Troubleshooting Hermes and LLMBase

Problem	Fix
Hermes returns an authentication error	Confirm that LLMBASE_CHAT_AGENT_KEY starts with llmbase_chat_... and that the base URL is https://llmbase.ai/api/v1/agents.
Hermes cannot use the selected model	Retrieve /api/v1/agents/models again and choose an exact returned model ID.
Hermes rejects a model at startup	Check the model context length. Hermes requires at least 64K tokens for its agent workflows.
A tool or image workflow fails	Verify the model currently supports the required capability before enabling the feature.
You need EU hosting	Select an applicable model marked EU-hosted in the current LLMBase catalog and keep its exact ID in the Hermes configuration.

FAQ: Hermes Agent with LLMBase

Can Hermes Agent use LLMBase?

Yes. Hermes supports custom OpenAI-compatible endpoints, and LLMBase provides a dedicated agent endpoint for compatible agents.

Which LLMBase API key works with Hermes?

Use an llmbase_chat_... chat-agent key with https://llmbase.ai/api/v1/agents. Do not use a direct inference key for this subscription-backed agent connection.

Does Hermes need a special model configuration?

Use an exact model ID from the Agent API, and choose a model with at least 64K context. For advanced work, verify its current tool, image, structured-output, and reasoning capabilities too.

For additional detail, see Agent integrations, agent setup examples, and models for agents.