Config Generation

LlamaFarm Config Generation Skill

Generate LlamaFarm configurations from natural language descriptions using internalized patterns.

Command Entry Point

/llamafarm:config "I want to analyze FDA regulatory PDFs"
/llamafarm:config "Add OCR support to my existing project"
/llamafarm:config "Create a multi-model setup with Universal Runtime and OpenAI"

When to Load

Activate this skill when:

• User runs /llamafarm:config command
• User asks to create a new LlamaFarm project
• User wants to modify existing configuration
• User describes what they want to build

Generation Workflow

1. Parse intent - New project or modification?
2. Identify use case - PDF, markdown, mixed, code?
3. Select pattern - Load appropriate pattern file
4. Generate config - Apply pattern to user needs
5. Validate - Run lf projects validate --config llamafarm.yaml
6. Present - Show config with CLI next steps

Intent Detection

New project indicators:

• "I want to...", "Create...", "Build...", "New project..."
• "Set up...", "Make a...", "Start a..."

Modification indicators:

• "Add...", "Change...", "Update...", "Remove..."
• "Switch...", "Modify...", "Enable..."

If modifying, read the existing llamafarm.yaml first.

Use Case Detection

Progressive Disclosure

Load these files based on the use case:

• pdf-pattern.md - PDF document configurations
• markdown-pattern.md - Markdown/text notes configurations
• mixed-pattern.md - Mixed format configurations
• code-pattern.md - Code analysis configurations
• large-docs-pattern.md - Large complex document configurations
• models.md - Model configuration templates

Minimal Valid Config

Every generated config must include:

version: v1
name: <project-name>
namespace: default

runtime:
  models:
    - name: default
      provider: <universal|ollama|openai>
      model: <model-id>
      default: true

prompts:
  - name: default
    messages:
      - role: system
        content: |
          <appropriate-system-prompt>

rag:
  databases:
    - name: main_db
      type: ChromaStore
      config:
        collection_name: documents
        persist_directory: ./data/main_db
      embedding_strategies:
        - name: default_embeddings
          type: UniversalEmbedder
          config:
            model: nomic-ai/nomic-embed-text-v2-moe
            dimension: 768
      retrieval_strategies:
        - name: basic_search
          type: BasicSimilarityStrategy
          config:
            top_k: 10
          default: true
      default_embedding_strategy: default_embeddings
      default_retrieval_strategy: basic_search

  data_processing_strategies:
    - name: default_processor
      parsers:
        - type: <appropriate-parser>
          file_include_patterns: ["<patterns>"]
          priority: 100
      extractors:
        - type: ContentStatisticsExtractor

datasets:
  - name: default_dataset
    database: main_db
    data_processing_strategy: default_processor

Presenting Results

After generating a config, show it with annotations and next steps:

Configuration Generated: <project-name>
=======================================

File: llamafarm.yaml

Key choices:
  - Provider: <provider> (<reason>)
  - Parser: <parser> (<reason>)
  - Extractors: <list> (<reason>)

Next steps:
  1. Review the generated llamafarm.yaml
  2. lf start
  3. lf datasets create -s default_processor -b main_db my_dataset
  4. lf datasets upload my_dataset ./path/to/files/*
  5. lf datasets process my_dataset
  6. lf chat "Ask a question"

Post-Generation CLI Workflow

# Validate
lf projects validate --config llamafarm.yaml

# Start services
lf start

# Create dataset
lf datasets create -s <strategy> -b <database> <dataset-name>

# Upload files
lf datasets upload <dataset-name> ./path/to/files/*

# Process
lf datasets process <dataset-name>

# Chat
lf chat "Your question here"