What it does

Nucleoid is a declarative runtime for TypeScript that builds a knowledge graph from your code. Every statement—whether it defines a class, sets a property, or asserts a rule—becomes a node in a “Logic Graph” that the runtime continuously reevaluates as new information arrives. It runs on Bun and exposes both an IDE and a chat interface for interacting with the system.

The interesting bit

The runtime doesn’t separate data from logic. Declare $Human.mortal = true and later instantiate new Human('Socrates'), and the mortality property propagates automatically through the graph. The system cites Kahneman’s “thinking, fast and slow” as inspiration: it aims to combine quick, context-driven responses with deliberate logical deduction from the same underlying representation.

Key highlights

• IPL-inspired declarative syntax; runtime tracks relationships between logic and data statements
• Built-in knowledge graph updates and reevaluates conclusions as new statements arrive (“plasticity”)
• Runs on Bun; distributed via npm as @nucleoidai/ide and @nucleoidai/expert
• Includes web-based chat interface at nucleoid.ai/chat
• Apache 2.0 licensed

Caveats

• The README is heavy on vision and light on API specifics; exact performance characteristics and graph scaling limits aren’t quantified
• “Neuro-Symbolic AI” framing promises neural network integration, but the actual neural component appears to be aspirational—the current implementation is symbolic AI with a graph store
• The TypeScript examples mix 'use declarative' and 'use imperative' pragmas without clarifying when each mode is required or what the runtime does differently under the hood

Verdict

Worth a look if you’re building expert systems, rule engines, or RAG pipelines where explainability matters and you want logic to propagate like data in a reactive store. Skip it if you need battle-tested graph databases or actual neural network integration today—the neural side of “neuro-symbolic” is more roadmap than reality.