Перейти к содержанию

🔄 Авто-синхронизация: из Discussion #1323 каждые 6 часов.

Gonka x Hermes Agent

Автор: @Dankosik · Категория: 🙌 Show and Tell · Создано: 2026-06-08 16:12 UTC · Обновлено: 2026-06-08 16:12 UTC

📝 Описание

GonkaGate x Hermes Agent

Set up local hermes-agent to use GonkaGate:

npx @gonkagate/hermes-agent-setup

If you use Hermes Agent and have a GonkaGate API key, this should be a short setup step, not a manual edit of ~/.hermes/config.yaml and ~/.hermes/.env.

The utility configures Hermes through the supported provider: custom path, stores the raw key in Hermes' .env, and only offers models that are live on GonkaGate and launch-qualified for Hermes.

Requirements

  • hermes-agent installed
  • Hermes Agent v2026.5.16 / v0.14.0 or newer
  • Node.js >=22.14.0
  • a GonkaGate API key
  • an interactive terminal
  • Linux, macOS, or WSL2

Public onboarding follows current GonkaGate Terms availability rules. It is not intended for users or entities in the United States of America or U.S. territories.

Run

Default interactive setup:

npx @gonkagate/hermes-agent-setup

Specific Hermes profile:

npx @gonkagate/hermes-agent-setup --profile work

After setup:

hermes

Optional real-request smoke test:

hermes chat -Q --max-turns 1 -q "Do not use tools. Reply exactly: GonkaGate smoke test OK"

The utility prints that smoke test at the end, but does not run it automatically. It sends one real model request.

What it writes

The utility resolves the active Hermes config paths with Hermes itself. With --profile, it uses the profile-specific paths.

It writes:

  • ~/.hermes/config.yaml
  • ~/.hermes/.env

The managed config shape is:

model:
  provider: custom
  base_url: https://api.gonkagate.com/v1
  default: <selected-model>
  api_key: ${GONKAGATE_API_KEY}

The raw key is stored in .env:

GONKAGATE_API_KEY=gp-...

The raw key is not written to config.yaml. The utility also avoids taking over unrelated OPENAI_API_KEY usage.

What happens during setup

  1. Check Node, terminal support, OS support, Hermes availability, and Hermes version.
  2. Resolve the active config.yaml and .env paths.
  3. Read existing config.yaml, .env, auth.json, and Hermes job state.
  4. Stop on conflicts the utility should not overwrite.
  5. Ask for the GonkaGate key through a hidden prompt.
  6. Fetch GET https://api.gonkagate.com/v1/models.
  7. Intersect the live catalog with checked-in Hermes qualification artifacts.
  8. Let the user choose a qualified live model.
  9. Show the planned file changes.
  10. Back up files, write config.yaml, then write .env.
  11. Roll back config.yaml if the .env write fails.

The utility does not accept a plain --api-key flag, so the key does not land in shell history or process lists.

Model selection

Models are not guessed from the live catalog alone. A model must be:

  • present in GonkaGate /v1/models
  • checked in as launch-qualified for Hermes Agent

Current allowlist:

  • moonshotai/kimi-k2.6 (recommended default)
  • minimaxai/minimax-m2.7
  • qwen/qwen3-235b-a22b-instruct-2507-fp8

Live-only models are ignored. Artifact-only models that are no longer live are ignored too.

What /v1/models proves

The setup request proves:

  • the key authenticates
  • GonkaGate returns a model catalog

It does not prove:

  • billing balance
  • quota
  • first billable Hermes request readiness

Use the optional hermes chat smoke test when you want runtime proof.

Current contract

  • package: @gonkagate/hermes-agent-setup
  • command: npx @gonkagate/hermes-agent-setup
  • installed bin: hermes-agent-setup
  • provider path: provider: custom
  • base URL: https://api.gonkagate.com/v1
  • managed config keys: model.provider, model.base_url, model.default, model.api_key
  • managed secret key: GONKAGATE_API_KEY

Non-goals

This is not a general Hermes installer. It does not:

  • replace upstream hermes setup
  • support Hermes versions older than v2026.5.16 / v0.14.0
  • accept arbitrary custom base URLs
  • migrate legacy endpoint settings such as OPENAI_BASE_URL, LLM_MODEL, root-level provider / base_url, or legacy custom_providers
  • edit shell profiles
  • write directly to auth.json credential pools
  • generate repository-local .env files
  • claim native Windows launch support
  • claim that /v1/models proves billing, quota, or first-request readiness

💬 Комментарии