flyto-ai

Natural language to executable automation. flyto-ai turns plain language descriptions into reusable YAML workflows using 412+ pre-built modules from flyto-core.

Why flyto-ai?

Traditional AI coding tools generate code that must be executed directly — non-deterministic and error-prone. flyto-ai takes a different approach:

• LLM never writes code — only selects modules and fills parameters
• Every execution produces a reusable YAML workflow
• Results are deterministic — same params always produce same results
• Zero-cost replays — save learned blueprints, skip LLM on reruns

Quick Start

Installation

pip install flyto-ai

Interactive Chat

flyto-ai

One-shot Execution

flyto-ai chat "scrape the top 10 posts from Hacker News"

YAML-only Mode (no execution)

flyto-ai chat "scrape example.com" --plan

Architecture

User Message
  → LLM (OpenAI / Anthropic / Ollama)
    → Tool Selection (412 flyto-core modules)
      → Module Execution (deterministic)
        → Results + YAML Workflow
          → Blueprint Learning (self-improving)

Key Components

Component	Purpose
Agent Core	Chat loop, tool dispatch, safety layers
LLM Providers	OpenAI, Anthropic, Ollama with failover
Claude Code Agent	AI-driven code fixes with verification
Memory	Persistent conversation memory with semantic search
Channels	Telegram, Discord, Slack integrations
Prompt System	Three-layer prompt architecture

Supported LLM Providers

Provider	Models	Best For
OpenAI	gpt-4o, gpt-4o-mini	Function calling
Anthropic	Claude Sonnet 4.5, Haiku	Reasoning + multi-step
Ollama	Qwen, Llama, Mistral	Local inference, no API costs

Providers can be chained with automatic failover on 429/5xx errors.