elith
(v0.1.0)
// advanced-usage [ 2.01 ]

Advanced Usage

Provider configuration, context engine tuning, and local LLMs.

Advanced Usage

// Multiple Provider Configuration

Configure multiple providers and switch between them:

# ~/.elith/config.toml

[providers.claude]
api_key = "sk-ant-..."
model = "claude-3-5-sonnet-20241022"
enabled = true

[providers.lmstudio]
base_url = "http://localhost:1234/v1"
model = "local-model"
enabled = true

[providers.openrouter]
api_key = "sk-or-..."
model = "anthropic/claude-3.5-sonnet"
enabled = false

Switch providers by enabling/disabling them in the config file.

// Context Engine Tuning

The context engine selects 4-6 files per task. You can influence selection by:

1.Obsidian Vault Integration

Create an Obsidian vault in your project to provide context:

your-project/
├── .obsidian/
├── docs/
│   ├── architecture/
│   │   └── system-design.md
│   ├── standards/
│   │   └── coding-standards.md
│   └── tasks/
│       └── current-sprint.md

The context engine reads these markdown files to understand:

  • Project architecture
  • Coding standards
  • Current priorities
  • Technical decisions
2.File Naming Conventions

The context engine prioritizes files with relevant names:

  • README.md, ARCHITECTURE.md - High priority
  • Files matching the task keywords
  • Recently modified files
  • Test files related to the target
3.Repository Structure

Organize code logically:

src/
├── auth/          # Authentication module
├── cache/         # Caching layer
├── api/           # API routes
└── utils/         # Utilities

Clear structure helps the context engine select relevant files.

// Using Local LLMs

LM Studio

  1. Download and install LM Studio
  2. Load a model (e.g., CodeLlama, DeepSeek Coder)
  3. Start the local server (default: http://localhost:1234)
  4. Configure Elith:
[providers.lmstudio]
base_url = "http://localhost:1234/v1"
model = "local-model"  # Use the model name from LM Studio
enabled = true
  1. Use Elith normally:
bashcopy
elith explain src/auth/login.ts

Recommended Local Models

For code tasks:

  • DeepSeek Coder 33B - Best for code generation
  • CodeLlama 34B - Good for explanations
  • Phind CodeLlama 34B - Optimized for coding

For general tasks:

  • Mistral 7B - Fast, good quality
  • Llama 2 13B - Balanced performance

// Session Management

View Session History

# List all sessions
ls -la bob-reports/

# View a specific session
cat bob-reports/session_20260518_123045_abc123.md

Session Log Format

Each session log contains:

# Elith Session Report

**Session ID:** abc123
**Started:** 2026-05-18T12:30:45Z
**Operation:** refactor
**Provider:** claude-3-5-sonnet-20241022

## Context
- Files analyzed: 6
- Total tokens: 8,234

## Conversation
[Full LLM conversation transcript]

## Skills Used
1. read_file: src/cache/redis-client.ts
2. write_file: src/cache/connection-pool.ts
3. search_code: pattern="redis.createClient"

## Results
- Files modified: 4
- Lines added: 127
- Lines removed: 89

// Service Management

Start/Stop Backend

# Start backend (auto-starts on first CLI use)
elith service start

# Check status
elith service status

# Restart backend
elith service restart

# Stop backend
elith service stop

Backend Configuration

The backend runs on http://localhost:8000 by default. It:

  • Auto-starts when you run any elith command
  • Runs in the background
  • Logs to ~/.elith/logs/
  • Can be managed via elith service commands

// Environment Variables

Override config with environment variables:

# Use a different provider for one command
ANTHROPIC_API_KEY=sk-ant-... elith explain src/app.ts

# Use a different model
ANTHROPIC_MODEL=claude-3-opus-20240229 elith architect "Design new feature"

# Use OpenRouter
OPENROUTER_API_KEY=sk-or-... elith refactor "Optimize database queries"

// Debugging

Enable Verbose Logging

# Set log level in config
[logging]
level = "DEBUG"

Check Backend Logs

tail -f ~/.elith/logs/backend.log

Test Provider Connection

# Test if provider is working
elith explain --test

// Performance Tips

  1. Keep Obsidian vault concise - Only include essential docs
  2. Use specific file paths - elith explain src/auth/login.ts is faster than elith explain src/
  3. Local LLMs - Faster for small tasks, no API costs
  4. Session cleanup - Periodically clean old sessions from bob-reports/

// Limitations

Current limitations (as of v0.1.0):

  • No parallel agent execution (single provider per task)
  • No approval gates (all changes are made immediately)
  • No webhooks or callbacks
  • No spend limits or token tracking
  • Context engine uses simple heuristics (not ML-based)

These features are planned for future releases.

// Next Steps