410979729/scope-recall-hermes

What it solves

Scope Recall provides a persistent, shared memory layer for AI agents. It solves the problem of fragmented agent memory by allowing multiple AI assistants—such as Hermes, Codex, Claude Code, WorkBuddy, and DeepSeek Harness—to access and contribute to a single, unified knowledge store. It ensures that information shared with one agent can be recalled by another, while maintaining strict control over what is stored as a "fact."

How it works

At its core, the project uses SQLite as the primary authority for storing verified facts, ensuring data integrity. It supports optional vector database companions (via LanceDB) to enable semantic search capabilities. The system acts as a middleware layer: it provides host adapters and an MCP (Model Context Protocol) server that agents use to read, write, and trace relationships between stored pieces of information. It explicitly avoids turning raw chat logs into facts, relying instead on explicit user confirmation to update or correct stored knowledge.

Who it’s for

This tool is designed for power users and developers who rely on multiple AI coding assistants or agents and want them to share a consistent, long-term memory of their preferences, project details, and past interactions across different sessions and tools.

Highlights

• Unified Memory: Allows different AI agents to share a single, synchronized memory store.
• Fact-Based Storage: Uses SQLite as the source of truth, ensuring that only confirmed information is treated as a fact.
• Agent-Operated Maintenance: Includes built-in skills that allow users to ask their agents to handle installations, upgrades, and migrations.
• Flexible Integration: Supports multiple host environments including Hermes, Codex, Claude Code, WorkBuddy, and DeepSeek Harness.
• Traceable Evidence: Provides a trace feature to query multi-hop relationships between stored facts without invoking additional models.