All plugins

wally

Wally by RunAnywhere as a Hermes model provider: open-weight models (DeepSeek, GLM, Kimi, MiniMax) on an OpenAI-compatible API, each with its context window, tool support and image input declared, and reasoning effort sent as Wally's top-level reasoning_effort, clamped to the levels each model accepts. Pick Wally in `hermes model` or use `--provider wally`. Disclosure — registers one model provider at import and nothing else (no tools, hooks, middleware, commands, background work or file access); chat requests (prompts, tool schemas and results, images for image-capable models) and the model list go to https://inference.runanywhere.ai/v1 (or WALLY_BASE_URL) through Hermes's own client with WALLY_API_KEY, which Hermes reads; usage is billed to your Wally account; the plugin itself opens no connections.

RunanywhereAIModelsv0.1.0Updated 1 hour ago

About

Wally by RunAnywhere as a Hermes model provider: open-weight models (DeepSeek, GLM, Kimi, MiniMax) on an OpenAI-compatible API, each with its context window, tool support and image input declared, and reasoning effort sent as Wally's top-level reasoning_effort, clamped to the levels each model accepts. Pick Wally in `hermes model` or use `--provider wally`. Disclosure — registers one model provider at import and nothing else (no tools, hooks, middleware, commands, background work or file access); chat requests (prompts, tool schemas and results, images for image-capable models) and the model list go to https://inference.runanywhere.ai/v1 (or WALLY_BASE_URL) through Hermes's own client with WALLY_API_KEY, which Hermes reads; usage is billed to your Wally account; the plugin itself opens no connections.

Wally for Hermes

Use Wally, RunAnywhere's OpenAI-compatible API for open-weight
models, as a model provider in Hermes Agent.
Requires Hermes 0.22.6 or later and a Wally API key from
console.runanywhere.ai. The plugin has no dependencies.

Connect

hermes plugins install wally
hermes model

Before the catalog listing lands, install from the repository instead:
hermes plugins install "https://github.com/RunanywhereAI/hermes-wally-provider#wally".

Select Wally, paste your WALLY_API_KEY when asked, pick a model, then a reasoning
level. Hermes saves the key and the choice for the active profile. One-off use:

hermes chat --provider wally -m deepseek-v4.1-flash

Models

All models support tool calling and reasoning. Values are from Wally's live catalog,
GET https://inference.runanywhere.ai/v1/models.

| Model | Context window | Image input |
|---|---:|---|
| deepseek-v4.1-flash (default) | 1,048,576 | Yes |
| glm-5.3-flash (background tasks) | 1,048,576 | No |
| kimi-k3 | 1,048,576 | Yes |
| deepseek-v4.1-flash-fast | 1,048,576 | Yes |
| glm-5.3-fast | 1,048,576 | No |
| deepseek-v4-pro-fast | 1,048,576 | No |
| deepseek-v4.1-flash-1017 | 1,048,576 | No |
| deepseek-v4-flash-0731 | 1,048,576 | No |
| minimax-m3 | 524,287 | No |

When the live catalog is reachable, hermes model lists it; the table above is the fallback.
Your account decides which models you can call. Prices are on
runwally.com/pricing; Hermes reads them from the catalog
to estimate session cost.

Reasoning

Hermes's reasoning level is sent as Wally's top-level reasoning_effort. Unset sends nothing,
so the model's default applies. GLM models cannot turn thinking off: for glm-5.3-flash
"off" sends minimal, for glm-5.3-fast it sends low. A level a model lacks becomes the
nearest weaker one (glm-5.3-fast accepts low, medium, high and max). Hermes's ultra
is sent as max.

Images you attach, and images inside tool results (screenshots, vision_analyze), go to
the models with image input as images. For text-only models Hermes describes them with a
vision model first, deepseek-v4.1-flash by default.

Configuration

| Variable | Purpose |
|---|---|
| WALLY_API_KEY | API key (required) |
| WALLY_BASE_URL | Optional endpoint override, default https://inference.runanywhere.ai/v1 |

The provider id is wally (alias runanywhere), e.g. wally:glm-5.3-flash.

Scope

Prompts, tool definitions and results, attached images, and model-list requests go to the
configured Wally endpoint through Hermes's own client, with the key Hermes reads. Usage is
billed to your Wally account. The plugin registers one model provider and nothing else: no
tools, hooks, background work, telemetry, file access or self-update. Update it with
hermes plugins update wally.

MIT license.

README from the reviewed commit 8a9a0dc; it updates when the author re-pins.