Hermes Command Code Provider

A Hermes provider bridge for using Hermes Agent with a Command Code Go subscription.

Why this exists

Command Code Go subscriptions can authenticate the Command Code CLI, but they do not include direct access to the Command Code provider API. This plugin works around that limitation for personal Hermes usage by running a local OpenAI-compatible bridge backed by your existing Command Code subscription login.

Chinese explanation: 为什么 Go 订阅可以这样使用.

This project provides:

• a Hermes model-provider plugin named commandcode
• a local OpenAI-compatible bridge at http://127.0.0.1:8788/v1
• model metadata with context_length so Hermes context usage displays correctly
• streaming usage chunks so Hermes updates token/session statistics
• tool-call translation between OpenAI Chat Completions and Command Code events
• reasoning event translation so thinking-only turns can continue correctly

Requirements

• macOS or Linux
• Hermes Agent v0.14.0 or newer
• curl
• Python 3
• a Command Code login or API key

Authenticate with Command Code first:

cmd login

Or export an API key:

export COMMANDCODE_API_KEY="..."

The bridge also reads ~/.commandcode/auth.json and ~/.pi/agent/auth.json.

Install

git clone https://github.com/MitoroMisaka/hermes-commandcode-provider.git
cd hermes-commandcode-provider
./install.sh --configure

install.sh copies the provider plugin, bridge, and helper command into your Hermes home. With --configure, it also updates ~/.hermes/config.yaml and backs it up first. It also applies small Hermes compatibility patches for Command Code:

• the model picker counts model-provider plugin fallback models correctly
• hermes doctor accepts Command Code slash-form model ids

Use

Start Hermes normally after configuration:

hermes

Or run a one-shot test:

hermes -z 'Reply with exactly OK.'

You can also use the helper, which starts the local bridge before running Hermes:

hermes-commandcode -z 'Reply with exactly OK.'

Switch models explicitly:

hermes --provider commandcode -m moonshotai/Kimi-K2.5 -z 'Reply with exactly OK.'
hermes --provider commandcode -m deepseek/deepseek-v4-flash -z 'Reply with exactly OK.'

Supported Models

The bridge exposes the live Command Code /provider/v1/models catalog when available, and falls back to a bundled catalog. Known context windows include:

Model	Context
moonshotai/Kimi-K2.6	256000
moonshotai/Kimi-K2.5	256000
deepseek/deepseek-v4-flash	1000000
deepseek/deepseek-v4-pro	1000000
claude-sonnet-4-6	1000000
claude-opus-4-7	1000000
gpt-5.4	400000
xiaomi/mimo-v2.5-pro	1000000
xiaomi/mimo-v2.5	1000000

Smoke Test

./scripts/smoke_test.sh

The smoke test checks:

• /v1/models returns model metadata and context lengths
• non-streaming chat works
• streaming chat returns a final usage chunk
• tool calls are translated correctly
• reasoning deltas are preserved as reasoning_content

Hermes Doctor Note

Some Hermes versions warn that slash-form model ids such as moonshotai/Kimi-K2.5 are "vendor-prefixed" and should belong to aggregators. Command Code legitimately uses these ids.

If your Hermes version shows that warning, either ignore it or run:

./scripts/patch_doctor.py

The patch only adds commandcode to the doctor allowlist for slash-form model ids.

Model Picker Note

Some Hermes versions auto-discover user-installed model-provider plugins but do not count their fallback_models on the first provider picker screen. In that case, Command Code can show as 0 models even though opening it displays the models and switching works.

Run this compatibility patch if needed:

./scripts/patch_model_picker.py

The patch only teaches the provider overview to count fallback models declared by installed model-provider plugins.

Files Installed

~/.hermes/commandcode_proxy.py
~/.hermes/plugins/model-providers/commandcode/__init__.py
~/.hermes/plugins/model-providers/commandcode/plugin.yaml
~/.local/bin/hermes-commandcode

No token is written into this repository. The bridge reads your existing Command Code auth at runtime.