CONSTELLATION-ENGINE / constellation-engine

What is constellation-engine?

Constellation Engine is a memory runtime for AI agents that treats long-term memory as an active process rather than passive storage. While most agent frameworks bolt on vector search and call it done, Constellation builds a living knowledge graph where each interaction spreads activation through typed nodes and weighted edges, waking related memories across turns. Built in JavaScript with Node.js 20+, it runs locally with SQLite, supports multiple LLM providers (Anthropic, OpenAI, Ollama, Gemini, and more), and ships with an Electron desktop shell for end users. The engine compiles activated graph material into structured context before handing it to the model, so the LLM receives a briefing rather than raw retrieved chunks.

Why is it gaining traction?

The hook is the shift from "retrieve similar documents" to "inject the aftermath of an activation event." Traditional RAG gives the model a stack of excerpts; Constellation gives it a cognitive briefing assembled from what the signal woke up. Features like Hebbian writeback, bi-temporal nodes, and post-turn consolidation address a real pain: agents that forget you the moment the tab closes. The model-agnostic design means you can swap providers without losing the agent's accumulated history, which is a genuine unlock for developers tired of vendor lock-in.

Who should use this?

Developers building long-lived AI agents who need persistent identity and memory that survives model swaps. Researchers experimenting with cognitive architectures and spreading activation models. Power users who want a local-first agent with inspectable memory topology rather than opaque context windows. Not yet ready for production teams needing battle-tested stability or enterprise support.

Verdict

Constellation Engine tackles a real problem with an interesting architectural approach, but the 0.699% credibility score and 31 stars reflect an early-stage project with limited community validation. The documentation is thorough for its scope, but test coverage and maturity markers are not yet established. Worth exploring as a research platform or for personal use, but wait for a stable release before betting production workloads on it.