hermes-provider-venice

Venice AI model provider plugin for Hermes Agent.

Venice has a privacy-first OpenAI-compatible inference API. This package registers a venice provider so Hermes can route chat through https://api.venice.ai/api/v1.

Disclaimer: This is an independent, open source project. It is not officially maintained, endorsed, or supported by Venice AI.

Install

# From PyPI:
pip install hermes-provider-venice

# From the latest GitHub main branch:
pip install "git+https://github.com/joshua-mo/hermes-provider-venice.git@main"

# Or from this checkout:
pip install .

For a reproducible GitHub installation, replace main with a release tag.

For Hermes releases without native packaged-provider discovery, or when the package is installed outside Hermes's environment, run the compatibility installer:

hermes-provider-venice install

The package exposes both the current hermes_agent.plugins entry point and the dedicated hermes_agent.model_providers entry point. Hermes releases with native packaged-provider discovery load it automatically when it is installed in the same Python environment as Hermes. The compatibility command copies the same profile to $HERMES_HOME/plugins/model-providers/venice.

To upgrade to the latest release later, run:

hermes-provider-venice update

This runs pip install --upgrade for the package in the current environment and, if the compatibility installer was used, refreshes the copy in $HERMES_HOME. Pass --pre to include pre-releases, or --source "git+https://github.com/joshua-mo/hermes-provider-venice.git@main" to update from GitHub. Restart Hermes afterwards to load the new version.

Add your API key (create one at https://venice.ai/settings/api):

# ~/.hermes/.env
VENICE_API_KEY=your-key-here

Use

hermes -z "hello" --provider venice -m zai-org-glm-5-2

Or in ~/.hermes/config.yaml:

model:
  provider: venice
  default: zai-org-glm-5-2

The live picker fetches GET /models?type=text and filters out offline, non-text, and non-tool-capable models. If that fails, these curated fallbacks are used (refreshed 2026-07-16 from the Venice catalog):

Full catalog: https://api.venice.ai/api/v1/models?type=text or https://docs.venice.ai

What the profile does

• auth_type: api_key via VENICE_API_KEY
• Optional VENICE_BASE_URL override
• Sets venice_parameters.include_venice_system_prompt: false so Hermes's system prompt is not appended to Venice's defaults
• Passes prompt_cache_key from the Hermes session id for better cache locality
• Filters the live catalog to online, text, tool-capable models
• Resolves each model's context window from Venice's live catalog
• Injects live per-token pricing from Venice's catalog into Hermes's cost tracking so session cost, the expensive-model guard, and usage reports work for Venice models (Venice reports per-million USD rates that Hermes's generic OpenAI-compatible pricing parser cannot read)
• Surfaces the same live pricing and per-model capabilities (reasoning, tools, vision, context) in Hermes's model pickers, including the desktop UI's price column and reasoning toggle
• Maps Hermes reasoning controls to Venice's recommended disable and model-supported effort parameters, clamping unsupported effort levels and omitting effort entirely for models that reason with a fixed budget

Develop

See DEVELOPMENT.md for setup, checks, local Hermes integration, and live API testing.

Contributions are described in CONTRIBUTING.md.

Uninstall

rm -rf ~/.hermes/plugins/model-providers/venice
# optionally: uv tool uninstall hermes-provider-venice