Gen3 Gt Api Runbooks Hermes - GT-Edge-AI/GT-AI-OS GitHub Wiki

Start Here

Point Hermes Agent at the tenant GT API with a first-class custom OpenAI-compatible endpoint (provider: custom, base_url, api_key); use a published agent alias as the model id after validating GET /v1/models. Recommended preset: Agentic IDE or editor client.

Why this matters

This runbook maps Hermes Agent to GT API routes operators actually publish.

Details

Compatibility: Native OpenAI-compatible ยท Category: Developer tooling

Official documentation

Configuration fields

  • model.provider: custom
  • model.base_url: https://<tenant-host>/api/tenant
  • model.api_key: gtak_...
  • model.default (or model.model): published-agent-alias
  • custom_providers[].api_mode: chat_completions

GT route mapping

GT route Verdict Client integration
GET /v1/models native Set model.default to published alias; optional GET /v1/models during setup
POST /v1/chat/completions native Hermes custom provider (api_mode: chat_completions) โ†’ /v1/chat/completions
POST /v1/embeddings not_supported Only if you add explicit embed calls in agent tooling
POST /v1/audio/transcriptions not_supported Not default Hermes OpenAI custom path
POST /v1/audio/speech not_supported Not default Hermes OpenAI custom path
POST /v1/images/generations not_supported Not default unless auxiliary vision routes to GT
POST /v1/conversations/files gt_extension Custom HTTP orchestration outside Hermes defaults
POST /v1/datasets/{id}/files gt_extension Custom HTTP orchestration outside Hermes defaults
GET /v1/files/{id} gt_extension Custom polling in agent tooling

Not supported in this product

  • Hermes auxiliary/vision routing may target a different provider unless you configure every task to use the GT custom endpoint.
  • Hermes does not manage GT conversation headers or dataset multipart uploads in the default custom-provider path.

Prerequisites

  • Hermes Agent installed (hermes setup or package install per Nous Research docs).
  • Publish at least one agent alias or raw-model catalog entry for agentic chat.
  • Create a dedicated agentic-editor inference key scoped to intended aliases.

Setup steps

  1. Run hermes model and choose Custom endpoint, or edit ~/.hermes/config.yaml.
  2. Set base URL to https:///api/tenant (Hermes appends /v1/chat/completions โ€” do not include a trailing /v1 segment unless your Hermes build expects the full path; use the tenant origin ending in /api/tenant).
  3. Set API key to a GT bearer token (gtak_...).
  4. Set the model name to a published alias from GET /v1/models.
  5. Send a short prompt in the Hermes terminal UI and confirm the response uses the GT-published model.

GT extensions and caveats

  • Hermes does not natively send X-GT-Conversation-Id or multipart dataset uploads โ€” add explicit HTTP only if you orchestrate GT file routes yourself.
  • Agent runners can generate sustained traffic; monitor quotas and keep alias allowlists narrow on the inference key.

Validation checklist

  • Hermes custom provider reaches GT API without 401/404 on chat.
  • Model id matches a published alias visible to that key.
  • Revoking or rotating the GT key stops Hermes traffic as expected.

Plain-text export

Hermes Agent runbook
Native OpenAI-compatible ยท Developer tooling
Recommended key preset: Agentic IDE or editor client
Evidence: documented compatibility (vendor docs cross-check)

Point Hermes Agent at the tenant GT API with a first-class custom OpenAI-compatible endpoint (`provider: custom`, `base_url`, `api_key`); use a published agent alias as the model id after validating `GET /v1/models`.

Official documentation:
- https://hermes-agent.nousresearch.com/docs/integrations/providers
- https://hermes-agent.nousresearch.com/docs/developer-guide/adding-providers
- https://github.com/NousResearch/hermes-agent

Configuration fields:
- model.provider: custom
- model.base_url: https://<tenant-host>/api/tenant
- model.api_key: gtak_...
- model.default (or model.model): published-agent-alias
- custom_providers[].api_mode: chat_completions

GT route mapping:
- GET /v1/models (native): Set `model.default` to published alias; optional `GET /v1/models` during setup
- POST /v1/chat/completions (native): Hermes custom provider (`api_mode: chat_completions`) โ†’ `/v1/chat/completions`
- POST /v1/embeddings (not_supported): Only if you add explicit embed calls in agent tooling
- POST /v1/audio/transcriptions (not_supported): Not default Hermes OpenAI custom path
- POST /v1/audio/speech (not_supported): Not default Hermes OpenAI custom path
- POST /v1/images/generations (not_supported): Not default unless auxiliary vision routes to GT
- POST /v1/conversations/files (gt_extension): Custom HTTP orchestration outside Hermes defaults
- POST /v1/datasets/{id}/files (gt_extension): Custom HTTP orchestration outside Hermes defaults
- GET /v1/files/{id} (gt_extension): Custom polling in agent tooling

Not supported in this product:
- Hermes auxiliary/vision routing may target a different provider unless you configure every task to use the GT custom endpoint.
- Hermes does not manage GT conversation headers or dataset multipart uploads in the default custom-provider path.

Prerequisites:
- Hermes Agent installed (`hermes setup` or package install per Nous Research docs).
- Publish at least one agent alias or raw-model catalog entry for agentic chat.
- Create a dedicated agentic-editor inference key scoped to intended aliases.

Setup steps:
1. Run `hermes model` and choose **Custom endpoint**, or edit `~/.hermes/config.yaml`.
2. Set base URL to https://<tenant-host>/api/tenant (Hermes appends /v1/chat/completions โ€” do not include a trailing /v1 segment unless your Hermes build expects the full path; use the tenant origin ending in /api/tenant).
3. Set API key to a GT bearer token (`gtak_...`).
4. Set the model name to a published alias from `GET /v1/models`.
5. Send a short prompt in the Hermes terminal UI and confirm the response uses the GT-published model.

GT extensions and caveats:
- Hermes does not natively send `X-GT-Conversation-Id` or multipart dataset uploads โ€” add explicit HTTP only if you orchestrate GT file routes yourself.
- Agent runners can generate sustained traffic; monitor quotas and keep alias allowlists narrow on the inference key.

Validation checklist:
- Hermes custom provider reaches GT API without 401/404 on chat.
- Model id matches a published alias visible to that key.
- Revoking or rotating the GT key stops Hermes traffic as expected.