@remnic/core

@remnic/core

Framework-agnostic memory and context engine for user-aware agents. Orchestration, storage, extraction, search, and trust zones -- inspectable and local by default.

Part of Remnic, open-source memory and context for user-aware agents.

Install

npm install @remnic/core

What it does

Remnic Core is the engine that powers scoped memory and context across AI agent sessions. It handles:

• Memory orchestration -- three-phase flow: recall before sessions, buffer during, extract after
• Storage -- plain markdown files with YAML frontmatter on your local filesystem
• Extraction -- GPT-5.2 or local LLM (Ollama, LM Studio) extracts durable knowledge from conversations
• Search -- hybrid BM25 + vector + reranking via QMD
• Trust zones -- namespace isolation and access control for multi-agent setups
• Entity tracking -- people, projects, tools, and their relationships
• Consolidation -- periodic merging, deduplication, and summarization
• Temporal supersession -- when a new fact writes a conflicting value for the same entityRef + structuredAttribute key, the older fact is marked status: "superseded" and excluded from recall by default. Opt in to audit/history via temporalSupersessionIncludeInRecall: true. Controlled by temporalSupersessionEnabled (default true). See issue #375.

Usage

Most users interact with Remnic through a higher-level package:

Use @remnic/core directly when building a custom integration or embedding the memory engine in your own agent framework.

import { Orchestrator } from "@remnic/core";

Fallback LLM

The core includes a fallback LLM client that resolves providers from your gateway config or OpenClaw's built-in provider catalog. It supports OpenAI-compatible and Anthropic APIs with automatic auth resolution and provider fallback chains.

License

MIT