hermes - 💡(How to fix) Fix [How-to] Volcengine Coding Plan configuration + model_normalize bug with vendor-prefixed model IDs

Context

Configuring Hermes Agent with Volcengine (火山引擎) Coding Plan has non-obvious traps. Sharing the working configuration and key findings.

Note: The original version of this issue contained errors about model name formats. This has been corrected after verification with another agent that independently solved the same problem.

Key Finding: Two Model Name Formats, Different Behaviors

Volcengine Coding Plan has two naming schemes that both work at the API level, but behave differently in Hermes:

The /model picker reads from the /models API endpoint, which only returns date-suffixed IDs. The dot-separated names exist as a compatibility layer but aren't listed in /models, so the picker can't find them.

Verdict: Use date-suffixed format if you want /model switching to work. Dot format only works as the model.default in config (since Hermes sends it directly to the API without picker validation).

Models to Avoid

Some models that work via dot notation are retiring or unlisted:

Working Configuration

model:
  default: volcengine/doubao-seed-2-0-pro-260215

custom_providers:
  - name: volcengine
    base_url: https://ark.cn-beijing.volces.com/api/coding/v3
    api_key: YOUR_KEY
    api_mode: chat_completions
    model: doubao-seed-2-0-pro-260215
  - name: volcengine
    base_url: https://ark.cn-beijing.volces.com/api/coding/v3
    api_key: YOUR_KEY
    api_mode: chat_completions
    model: doubao-seed-2-0-code-preview-260215
  - name: volcengine
    base_url: https://ark.cn-beijing.volces.com/api/coding/v3
    api_key: YOUR_KEY
    api_mode: chat_completions
    model: doubao-seed-2-0-mini-260215
  - name: volcengine
    base_url: https://ark.cn-beijing.volces.com/api/coding/v3
    api_key: YOUR_KEY
    api_mode: chat_completions
    model: doubao-seed-2-0-lite-260215
  - name: volcengine
    base_url: https://ark.cn-beijing.volces.com/api/coding/v3
    api_key: YOUR_KEY
    api_mode: chat_completions
    model: doubao-seed-code-preview-251028
  - name: volcengine
    base_url: https://ark.cn-beijing.volces.com/api/coding/v3
    api_key: YOUR_KEY
    api_mode: chat_completions
    model: deepseek-v3-2-251201
  - name: volcengine
    base_url: https://ark.cn-beijing.volces.com/api/coding/v3
    api_key: YOUR_KEY
    api_mode: chat_completions
    model: glm-4-7-251222

Each model is a separate entry with the same name. Hermes auto-groups them in the /model picker.

Verified Model List (2026-04)

model_normalize Note (SiliconFlow specific)

Separate issue: model_normalize.py strips / from custom provider model names (deepseek-ai/DeepSeek-V3.2 → DeepSeek-V3.2), breaking SiliconFlow which requires vendor/model format. This is unrelated to Volcengine but affects users combining Volcengine + SiliconFlow in the same config.

Reference

• https://www.volcengine.com/docs/82379/1925114 — Official Coding Plan guide
• https://developer.volcengine.com/articles/7615528054736945158 — OpenClaw + Coding Plan guide (Chinese)
• #7916 — custom_providers model field propagation fix