1908dd09fc
Ollama sizes a model's real context window by free VRAM at load time, often far below the GGUF trained max that /api/show reports, and its OpenAI-compatible endpoint has no options passthrough — per-request num_ctx and keep_alive are silently dropped, so the window cannot be controlled from the client. The compressor was being sized to the trained max (e.g. 262K for a model actually running at 32K). - Add query_ollama_loaded_context() reading the effective window from /api/ps (60s cache, never persisted — transient load state). - Reconcile after each successful response via sync_ollama_loaded_context(): resize the compressor to the loaded window and warn when it is below the tool-use minimum. Selection surfaces keep showing the trained max; explicit model.context_length still wins. No-op for non-Ollama providers. - Refresh model.ollama_keep_alive through the native API (rate-limited /api/generate ping) since /v1 drops it. - Remove the inert num_ctx/keep_alive request-body plumbing; warn that model.ollama_num_ctx has no effect and point at OLLAMA_CONTEXT_LENGTH / Modelfile num_ctx. Keep detection for the pre-flight window check. - Disable thinking on Ollama thinking models via reasoning_effort 'none' — the only switch its /v1 handler parses (think is dropped). |
||
|---|---|---|
| .. | ||
| alibaba | ||
| alibaba-coding-plan | ||
| anthropic | ||
| arcee | ||
| azure-foundry | ||
| bedrock | ||
| copilot | ||
| copilot-acp | ||
| custom | ||
| deepinfra | ||
| deepseek | ||
| fireworks | ||
| gemini | ||
| gmi | ||
| huggingface | ||
| kilocode | ||
| kimi-coding | ||
| minimax | ||
| nous | ||
| novita | ||
| nvidia | ||
| ollama-cloud | ||
| openai-codex | ||
| opencode-zen | ||
| openrouter | ||
| qwen-oauth | ||
| stepfun | ||
| vertex | ||
| xai | ||
| xiaomi | ||
| zai | ||
| README.md | ||
Model Provider Plugins
Each subdirectory is a self-contained provider profile plugin. The
directory layout mirrors plugins/platforms/:
plugins/model-providers/
├── openrouter/
│ ├── __init__.py # registers the ProviderProfile
│ └── plugin.yaml # manifest: name, kind, version, description
├── anthropic/
│ ├── __init__.py
│ └── plugin.yaml
└── ...
How discovery works
providers/__init__.py._discover_providers() scans this directory (and
$HERMES_HOME/plugins/model-providers/) the first time anything calls
get_provider_profile() or list_providers(). Each __init__.py is
imported and expected to call providers.register_provider(profile).
User plugins at $HERMES_HOME/plugins/model-providers/<name>/ override
bundled plugins of the same name — last-writer-wins in
register_provider(). Drop a file there to replace a built-in.
Adding a new provider
-
Create
plugins/model-providers/<your_provider>/__init__.py:from providers import register_provider from providers.base import ProviderProfile my_provider = ProviderProfile( name="your-provider", aliases=("alias1", "alias2"), display_name="Your Provider", description="One-line description shown in the setup picker", signup_url="https://your-provider.example.com/keys", env_vars=("YOUR_PROVIDER_API_KEY", "YOUR_PROVIDER_BASE_URL"), base_url="https://api.your-provider.example.com/v1", default_aux_model="your-cheap-model", ) register_provider(my_provider) -
Create
plugins/model-providers/<your_provider>/plugin.yaml:name: your-provider-profile kind: model-provider version: 1.0.0 description: Short sentence about the provider author: Your Name
Nothing else needs to change. auth.py, config.py, models.py,
doctor.py, model_metadata.py, runtime_provider.py, and the
chat_completions transport all auto-wire from the registry.
Non-trivial profiles
Override the ProviderProfile hooks in a subclass for per-provider
quirks — see plugins/model-providers/openrouter/__init__.py for
build_extra_body and build_api_kwargs_extras examples, and
plugins/model-providers/gemini/__init__.py for thinking_config
translation.