explaingit

410979729/scope-recall

Analysis updated 2026-06-24

33PythonAudience · developerComplexity · 3/5LicenseSetup · moderate

TLDR

A memory plugin for the Hermes AI agent framework that stores facts in SQLite with a LanceDB vector companion and applies per-scope rules so notes do not leak across chats.

Mindmap

mindmap
  root((scope-recall))
    Inputs
      User preferences
      Project notes
      Turn captures
    Outputs
      Bounded recall prompt
      Audit trail rows
    Use Cases
      Persistent agent memory
      Scope-isolated notes
      Offline embeddings
    Tech Stack
      Python
      SQLite
      LanceDB
      Hermes
Click or tap to explore — scroll the page freely

Code map

Detail Auto

An interactive map of this repo's files and how they connect — its source is parsed live in your browser. Click Visualize to build it.

filefunction / class

What do people build with it?

USE CASE 1

Give a Hermes-based agent persistent memory that survives new chat windows

USE CASE 2

Prevent scratch notes from one group chat leaking into another

USE CASE 3

Run agent memory offline using the deterministic local-hash embedding fallback

USE CASE 4

Audit and rebuild an agent's memory from the SQLite authoritative store

What is it built with?

PythonSQLiteLanceDBHermes

How does it compare?

410979729/scope-recallarahim3/mlx-dsparkgongyichuren/tg-watchbot
Stars333333
LanguagePythonPythonPython
Setup difficultymoderateeasymoderate
Complexity3/53/53/5
Audiencedeveloperdeveloperdeveloper

Figures from each repo's GitHub metadata at analysis time.

How do you get it running?

Difficulty · moderate Time to first run · 30min

Requires Python 3.11+ and a working Hermes install, embeddings need a Gemini-compatible API key unless the local hash fallback is acceptable.

MIT license: use, modify, and redistribute freely including for commercial use, as long as you keep the copyright notice.

In plain English

scope-recall is a memory plugin for Hermes, an AI agent framework from Nous Research. The job of a memory plugin is to decide what an AI assistant should remember between conversations and what it should not. This one is built around a specific complaint the README highlights: most agent memory tools either forget too much, so opening a new chat window means losing everything, or share too much, so a temporary note left in one group conversation leaks into another. The design splits memory into two layers and applies different rules to each. The first layer is a SQLite database treated as the authoritative truth store: every record lives here as plain rows that can be audited and rebuilt. The second is a LanceDB vector store, used as a companion for semantic search. LanceDB is rebuildable from the SQLite content, so if it gets corrupted or out of sync, it can be regenerated. Retrieval is hybrid, combining SQLite full-text search candidates with LanceDB semantic candidates, then rendering a bounded prompt. The scope rules are the heart of the project. Facts tagged as user preferences, explicit memories, project notes, or ops notes are shared durably across chat windows for the same user-plus-agent identity, so a decision made in one conversation can be recalled in the next. Raw general turn captures, the model's scratch notes from a single conversation, stay local to that chat, thread, or session and do not bleed into other contexts. Recall is current-turn scoped, meaning the plugin searches only for memories relevant to the active query, not for whatever was in the previous turn's buffer. Cross-scope governance actions like export, dedupe, or repair fail closed unless an explicit maintenance mode is set. Embeddings can run through a hosted OpenAI-compatible Gemini endpoint when an API key is configured, with a deterministic local-hash fallback so the plugin still works offline without keys. Installation is the standard Hermes plugin flow: clone into the plugins directory under HERMES_HOME, pip install in editable mode, set memory.provider to scope-recall in the Hermes config, and verify with hermes memory status. Python 3.11 or newer is required. An optional companion plugin called turn-closure-audit handles after-the-fact review of completed turns. The license is MIT.

Copy-paste prompts

Prompt 1
Install scope-recall into my Hermes setup and walk me through verifying it with hermes memory status.
Prompt 2
Adapt scope-recall's two-layer SQLite plus LanceDB design for a non-Hermes agent I am building in LangGraph.
Prompt 3
Show me how to add a new scope rule to scope-recall so meeting notes are shared across chats but never with other users.
Prompt 4
Write a test that proves scope-recall does not leak general turn captures across thread IDs.

Frequently asked questions

What is scope-recall?

A memory plugin for the Hermes AI agent framework that stores facts in SQLite with a LanceDB vector companion and applies per-scope rules so notes do not leak across chats.

What language is scope-recall written in?

Mainly Python. The stack also includes Python, SQLite, LanceDB.

What license does scope-recall use?

MIT license: use, modify, and redistribute freely including for commercial use, as long as you keep the copyright notice.

How hard is scope-recall to set up?

Setup difficulty is rated moderate, with roughly 30min to a first successful run.

Who is scope-recall for?

Mainly developer.

Open on GitHub → Explain another repo

This repo across BitVibe Labs

Verify against the repo before relying on details.