Models and Providers

HelseCLI is designed to be model-agnostic. You can switch between top-tier models from various providers seamlessly.

HelseCLI is engineered for model-agnosticism, enabling seamless transitions between leading Large Language Models from diverse providers.

Integrated Providers

Standard for software engineering and general computational reasoning.

Configuration: Requires OPENAI_API_KEY in the environment configuration.
Primary Model: gpt-4o (Optimized for balance of latency and intelligence).
Reasoning Models: o1-preview, o3-mini (Specifically for complex logical synthesis).

Recognized for high-fidelity technical writing and nuanced contextual analysis.

Configuration: Requires ANTHROPIC_API_KEY in the environment configuration.
Primary Model: claude-3-5-sonnet (Industry-leading performance in coding and research).

Optimized for high-throughput tasks and extensive context window requirements.

Enables private, air-gapped execution with zero token-based costs.

Configuration: Requires an active local Ollama instance. Default connection via localhost:11434.
Supported Models: llama3, codellama, qwen2.5-coder.

The active inference model can be dynamically reconfigured during any active session:

Command	Action
`!m`	Access the indexed model selection interface.
`!model <name>`	Explicitly transition to a defined model (e.g., `!model gpt-4o`).
`!default <name>`	Persist a model choice as the default for future initializations.

HelseCLI supports third-party OpenAI-compatible endpoints (e.g., LM Studio, LocalAI, Groq).

To integrate a custom endpoint, modify the custom_models.json manifest in the application root:

{
  "custom-inference-endpoint": {
    "api_base": "https://api.example-provider.com/v1",
    "provider": "openai",
    "is_custom": true
  }
}

Software Development: gpt-4o and claude-3-5-sonnet are recommended for code synthesis.
Large-Scale Document Analysis: gemini-1.5-pro is recommended for its extensive context window.
Iterative Research: gpt-4o is optimized for complex tool orchestration and web data synthesis.
Privacy-Restricted Tasks: Local execution via llama3 (Ollama) is recommended for data sensitive workflows.

Architecture

Tools System