Hermes Agent

Hermes Agent is Nous Research's open-source agent. It has the smoothest Cobble setup of any tool: its wizard verifies your endpoint against our live catalog, so every model on your plan appears automatically — no manual model lists.

Works best with: Agent Runner (pinned slot keeps long-running agents at consistent latency) or Coding.

Option A: setup wizard (recommended)

1. Install and launch Hermes. When it prompts for an inference provider, choose Custom OpenAI-compatible endpoint.
2. Base URL: https://api.cobble.network/v1
3. API key: your Cobble key from the dashboard.

Hermes calls /v1/models, lists what your key can access, and asks you to confirm a default. Pick qwen/qwen3.5-122b-a10b for general agent work — done.

Option B: manual config

In your Hermes config.yaml:

model:
  default: qwen/qwen3.5-122b-a10b
  provider: cobble

providers:
  cobble:
    base_url: https://api.cobble.network/v1
    key_env: COBBLE_API_KEY
    type: openai

Use key_env rather than an inline api_key — it keeps your key out of the config file. Leave api_mode at its default (chat_completions); that's what Cobble speaks.

Troubleshooting

• Wizard can't verify the endpoint — test the key directly: curl https://api.cobble.network/v1/models -H "Authorization: Bearer $COBBLE_API_KEY". A 401 means the key; a timeout means network/proxy.
• Model list looks short — Hermes shows what your plan includes. Models above your tier are available via wallet spending at API rates.
• 429 during long runs — unattended agents can exhaust a plan window. Set a wallet spend cap in the dashboard so runs continue at API rates instead of stopping.

Plan notes

Agent Runner's pinned slot is built for exactly this: a Hermes agent running for hours gets consistent latency instead of re-queueing per request. If your agent fires parallel tool calls, the Coding plan's 2 slots remove the serialization.