Claws & Code

Claws & Code (Claws and Effect) 🦞⚡

Transforming programming from typing code into orchestrating cognitive processes.

Inspired by the shift towards “Agentic Engineering,” this project implements a highly resilient Orchestrator-Worker paradigm. It features an event-driven triage system, adversarial quality gates, and a multi-provider “Bring Your Own Key” (BYOK) fallback architecture.

View on GitHub

🌟 Core Architecture

1. The Orchestrator (“Claw”)

The “Cognitive Brain.” It receives a high-level English objective and decomposes it into a dynamic graph of tasks. Instead of a rigid DAG, the Claw uses a Triage Router to handle complex, reactive workflows where agents trigger each other dynamically based on real-world events.

2. The Workers (“Code Instances”)

The “Specialized Hands.” Each worker executes a Tenacity Loop (Research -> Strategize -> Act -> Verify). They are ephemeral and decoupled; they simply perform their task and emit domain events (e.g., env:ready, ui:scaffolded) to trigger downstream logic.

3. Adversarial Quality Gates (Reviewers)

Inspired by “Pair Programming” philosophies, every sub-task must pass a dedicated Reviewer Agent before finalization. This adversarial loop ensures that hallucinations or edge cases are caught and sent back for revision before they cascade into the broader system.

4. Event-Driven Triage Router (Pub/Sub)

A fully reactive event broker that allows for massive decoupling and infinite branching. Agents subscribe to specific event patterns. When a “Worker” finishes, it publishes an event, instantly waking up the “Reviewer” or the next relevant “Worker” in the pipeline.

5. Multi-Provider BYOK with Fallback

A resilient, zero-persistence security model. Users provide their own API keys (Google, OpenAI, Anthropic) directly in the UI.

🖥️ The Perceptive UI: “The Swimlane Matrix”

When managing parallel AI agents, a linear chat feed is the wrong UX. We built a bespoke, high-density, industrial-cybernetic interface focusing on Parallelism.

🚀 Getting Started

This project is built for extreme speed and low overhead using FastAPI and uv.

# Clone the repository
git clone https://github.com/inventcures/claws-and-effect.git
cd claws-and-effect

# Create a virtual environment and install dependencies
uv venv
source .venv/bin/activate
uv pip install -r requirements.txt

# Start the Perceptive UI Server
uvicorn main:app --host 0.0.0.0 --port 8000