⚡️ LAO: Local AI Workflow Orchestrator

Chain. Build. Run. All offline. LAO is how developers bend AI to their will—no cloud, no compromise.

---

🧠 What is LAO?

LAO is a cross-platform desktop tool for chaining local AI models and plugins into powerful, agentic workflows. It supports prompt-driven orchestration, visual DAG editing, and full offline execution.

---

✨ Features

• Modular plugin system (Rust, local-first, dynamic loading)
• Offline DAG engine (retries, caching, lifecycle hooks)
• Prompt-driven agentic workflows (LLM-powered, system prompt file)
• Visual workflow builder (egui-based native GUI, drag & drop)
• CLI interface (run, validate, prompt, validate-prompts, plugin list)
• Prompt library (Markdown + JSON, for validation/fine-tuning)
• Test harness for prompt validation
• End-to-end execution from UI (execute and show logs/results)
• UI streaming run with real-time step events and parallel execution
• Node/edge editing in UI (drag, connect, edit, delete)
• Cross-platform support (Linux, macOS, Windows)
• Conditional/branching steps (output-based conditions)
• Multi-modal input (files, voice, images, video)
• Automated packaging (deb, rpm, AppImage, dmg, msi, zip)
• CI/CD pipeline (GitHub Actions, automated releases)
• Plugin explainability (lao explain plugin <name>)
• Plugin marketplace/discovery
• Live workflow status/logs in UI

---

🚀 Quickstart

GUI (Recommended)

# Run the native GUI with visual workflow builder
cargo run --bin lao-ui

CLI

# Run workflows from command line
cargo run --bin lao-cli run workflows/test.yaml

# Generate workflows from natural language
cargo run --bin lao-cli prompt "Summarize this audio and tag action items"

# Validate prompt library
cargo run --bin lao-cli validate-prompts

Build Plugins

# Build all plugins for your platform
bash scripts/build-plugins.sh

---

🧩 Prompt-Driven Workflows

LAO can generate and execute workflows from natural language prompts using a local LLM (Ollama). The system prompt is editable at core/prompt_dispatcher/prompt/system_prompt.txt.

Example:

lao prompt "Refactor this Python file and add comments"

---

📚 Prompt Library & Validation

• Prompts and expected workflows: core/prompt_dispatcher/prompt/prompt_library.md and .json
• Validate with: cargo run --bin lao-cli validate-prompts
• Add new prompts to improve LLM output and test new plugins

---

🛠️ Contributing Plugins & Prompts

• Add new plugins by implementing the LaoPlugin trait, building as a cdylib, and placing the resulting library in the plugins/ directory
• Expose a C ABI function named plugin_entry_point that returns a Box<dyn LaoPlugin>
• Add prompt/workflow pairs to the prompt library for validation and LLM tuning
• See docs/plugins.md and docs/workflows.md for details

---

📄 Documentation

• Architecture: docs/architecture.md
• Plugins: docs/plugins.md
• Workflows: docs/workflows.md
• CLI: docs/cli.md
• Observability: docs/observability.md

---

🌌 Manifesto

Cloud is optional. Intelligence is modular. Agents are composable.
LAO is how devs build AI workflows with total control.
No tokens. No latency. No lock-in.

Let’s define the category—one plugin at a time.