WeaveMindAI/weft

What it solves

Weft addresses the complexity of building AI systems that require coordination between LLMs, humans, APIs, and infrastructure. It eliminates the "plumbing" typically required to manage state, webhooks, and polling when integrating these disparate components into a single workflow.

How it works

Weft is a specialized programming language where AI and infrastructure primitives are first-class citizens. It uses a node-based architecture where components (like LLMs, databases, and communication tools) are represented as nodes that are wired together.

Key technical features include:

• Durable Execution: Powered by Restate, allowing programs to survive crashes and restarts, which is essential for long-running processes like waiting for human approval.
• Strong Typing: A compiler that validates connections, generics, and type mismatches before execution.
• Dual Representation: Programs can be edited as dense code for developers or as a visual graph for humans, with changes in one view automatically updating the other.
• Recursive Folding: Complex systems of nodes can be collapsed into single blocks to maintain high-level readability.

Who it’s for

AI builders and developers who want to build complex, multi-step AI agents or workflows that integrate human-in-the-loop interactions and durable state management without writing extensive boilerplate code.

Highlights

• First-class human interaction: Pause and resume programs via human query nodes without manual state management.
• End-to-end type safety: Compiler-level checks for architecture and connection validity.
• Built-in node catalog: Ready-to-use nodes for LLMs, Postgres, Discord, Slack, Telegram, WhatsApp, Email, and web search.
• Automatic visual graphing: Every program is automatically rendered as a graph for easier visualization.