Configuration Guide

HelseCLI is highly configurable to suit your workflow. Configuration is handled through environment variables, JSON files, and YAML files.

Configuration Reference

HelseCLI utilizes a multi-layered configuration system to manage system behavior, API integration, and project-specific parameters.

Environment Configuration (.env)

The .env file, located in the application root, manages secure credentials and global environmental flags.

API Integration Keys

VariableDescription
OPENAI_API_KEYPrimary credential for OpenAI services.
ANTHROPIC_API_KEYPrimary credential for Anthropic (Claude) services.
GOOGLE_API_KEYPrimary credential for Google Vertex AI / Gemini services.
OLLAMA_BASE_URLEndpoint for local inference via Ollama (Default: http://localhost:11434).

Feature-Specific Credentials

VariableDescription
TAVILY_API_KEYCredential for Tavily hyper-search integration.
BRAVE_API_KEYCredential for Brave Search API.
SERPER_API_KEYCredential for Serper.dev Google Search abstraction.
E2B_API_KEYCredential for E2B Cloud sandboxed execution.
STABILITY_API_KEYCredential for Stability AI image synthesis.

System Runtime Settings

VariableDescriptionDefault
HELSE_HOME_ENABLEDEnables the interactive home interface.1
HELSE_DEBUGActivates verbose diagnostic logging.0
HELSE_DEFAULT_MODELSets the system-wide default inference model.gpt-4o

Configuration Schemas

1. Global System Configuration (~/.helsecli_config.json)

Maintains persistent user preferences across different directory contexts.

  • Default inference parameters
  • Interface theme specifications
  • Interaction history retention policies

2. Local Project Configuration (.helsecli.yaml)

Enables repository-specific overrides and logic tuning.

project_name: "ResearchProjectAlpha"
description: "Statistical analysis and document synthesis workflow"
custom_instructions:
  - "Enforce PEP8 compliance for all Python generation"
  - "Utilize type-hinting for all function definitions"
verification:
  enabled: true
  linter: "pylint"

Contextual Instructions

HelseCLI integrates instructions from three primary levels:

  1. Global Layer: Defined within the system-wide configuration file.
  2. Project Layer: Resides in .helsecli.yaml or an AGENTS.md manifest.
  3. Session Layer: Volatile instructions injected during runtime via the /init command.

Knowledge Injection via AGENTS.md

Users can deploy an AGENTS.md file in the project root to provide comprehensive architectural context and standard operating procedures for the AI agent.

Model Selection Protocol

The inference model can be modified during an active session using the !m command. This opens the model selection interface, allowing for dynamic switching between configured providers.

Execution Sandboxing

HelseCLI supports three distinct execution environments for code validation:

  • Local Host: Direct execution (use with caution).
  • Docker: Local containerized isolation.
  • E2B: Managed cloud-based sandboxing for maximum security.

These environments are toggled via project configuration or environment-specific variables.