Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 12 additions & 2 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -14,12 +14,14 @@ python3 model-providers/cloudgpu/connect.py --spend-limit 5

Open the link, sign in with Google, review the cap and approve. If needed, top up on the owner page. The helper writes the returned key to the private shared CloudGPU credential store and, by explicit invocation of this Hermes helper, the private local `~/.hermes/.env`. It never prints the key or makes a payment. The $5 is a key spending cap, not an automatic charge. Pending approval survives restarts. Restart Hermes after successful connection; revocation and limits remain enforced by CloudGPU.

Do not install an unrelated similarly named package while the candidate SDK is not published. Existing manual key setup below remains available. Hermes framework compatibility still needs actual installed-Hermes acceptance; a local wrapper test is not end-to-end certification.
Do not install an unrelated similarly named package while the candidate SDK is not published. Existing manual key setup below remains available. Use an immutable `--ref <commit>` when installing a reviewed release. Runtime installation/registration acceptance is documented below; a new owner Google payment still requires the owner browser flow.

## Existing manual setup

```bash
hermes plugins install cloudgpu-provider && export CLOUDGPU_API_KEY=cgw-sk-... && hermes chat --provider cloudgpu -m deepseek-v4.1-flash
hermes plugins install cloudgpu-app/hermes-cloudgpu --enable
# Then connect through the reviewed Python SDK, or configure your private Hermes env.
hermes chat --provider cloudgpu -m deepseek-v4.1-flash
```

Create the key in the cloudgpu.app console. Put it in `~/.hermes/.env` to keep it; the installer also asks for it.
Expand Down Expand Up @@ -55,3 +57,11 @@ hermes plugins validate --install-deps model-providers/cloudgpu
## License

MIT.

## Repository layout

The repository root contains the native Hermes `plugin.yaml` and provider entry, so `hermes plugins install cloudgpu-app/hermes-cloudgpu` discovers a real plugin. The nested `model-providers/cloudgpu` copy remains for catalog compatibility; both entries must stay identical. The connector honors `HERMES_HOME` in the reviewed Python SDK and never writes keys into this repository.

## Runtime acceptance (2026-11-13)

Tested the actual upstream Hermes CLI/provider registry at revision `5c1abd257d0cc1b531ffee6e50bef0621aa49dc4` with Python3.12. Repository installation, native provider registration and live public model catalog (42 models at observation) passed. The connection helper resumed real production pending state in a new process, completed existing-owner approval, and wrote0600 credentials only in an isolated HERMES_HOME. The internal test key was revoked afterward; no GPU or model-generation charge occurred. This does not certify every Hermes version, a model conversation, a new Google login/payment, or a PyPI publication.
33 changes: 33 additions & 0 deletions __init__.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,33 @@
"""cloudgpu.app model provider for Hermes Agent.

Hermes drives models served by cloudgpu.app over the OpenAI-compatible API at
https://cloudgpu.app/v1 with a CLOUDGPU_API_KEY. The model list comes live
from GET /v1/models (public, no key needed); fallback_models is the static
list Hermes shows when that call fails.
"""

from providers import register_provider
from providers.base import ProviderProfile

cloudgpu = ProviderProfile(
name="cloudgpu",
aliases=("cloudgpu-app", "cloudgpu.app"),
display_name="CloudGPU",
description="cloudgpu.app: DeepSeek, Kimi, GLM and MiniMax on one OpenAI-compatible API",
signup_url="https://cloudgpu.app/api",
env_vars=("CLOUDGPU_API_KEY", "CLOUDGPU_BASE_URL"),
base_url="https://cloudgpu.app/v1",
auth_type="api_key",
default_aux_model="deepseek-v4.1-flash",
# [0] is the setup default; the live catalog at /v1/models lists the rest.
fallback_models=(
"deepseek-v4.1-flash",
"deepseek-v4-pro",
"kimi-k3",
"kimi-k2.7-code",
"glm-5.2",
"minimax-m3",
),
)

register_provider(cloudgpu)
11 changes: 11 additions & 0 deletions plugin.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,11 @@
name: cloudgpu-provider
kind: model-provider
version: 1.0.0
description: cloudgpu.app models (DeepSeek, Kimi, GLM, MiniMax) via the OpenAI-compatible API
author: cloudgpu.app
requires_hermes: ">=0.21.5"
requires_env:
- name: CLOUDGPU_API_KEY
description: cloudgpu.app API key (cgw-sk-...)
url: https://cloudgpu.app/api
secret: true