Scope Recall plugin
A shared local SQLite memory for Claude Code, Codex and Hermes: every fact has a source and can be corrected or deleted
- 542 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Hooks read and store whole agent conversations, including pasted code and keys
- With model routes configured, conversation fragments go to an external provider
- A false fact planted in a conversation can reach recall: suspicious sources are held, but the final check stays with you
Install
- Works in
Manual install
python -m pip install -U "hermes-scope-recall[codex]"Install into the Python environment the hooks will run from. Then scope-recall setup and confirm the plan.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Hooks read and store whole agent conversations, including pasted code and keys
- With model routes configured, conversation fragments go to an external provider
- A false fact planted in a conversation can reach recall: suspicious sources are held, but the final check stays with you
This is third-party code. Review the repository files before installing.
What it does
Scope Recall keeps agent conversations in a local SQLite database and pulls what matters into later sessions. The original source stays untouched, and facts and vectors that models derive from it trace back to the source quote and can be rebuilt. Claude Code, Codex, Hermes, WorkBuddy and DeepSeek Harness share one store: each fact carries the agent it came through, and a deletion through any of them applies to all. Automatic recall adds zero to six items that clear a lexical coverage or vector similarity floor. Acknowledgments, approvals and status prompts start no search, and unverified items never ship. The `scope-recall memory` command browses and searches facts, corrects and mutes them, and deletes a conversation after listing the dependent sources. `scope-recall ui` opens a local management page with Chinese and English interfaces. `scope-recall private on` stops keeping a session, a directory or the whole store, optionally for a set time. Before indexing, text is screened for embedded instructions, exfiltration links and hidden Unicode, and suspicious sources are held for manual review.
Who it is for. People who run long projects across several agents and want a shared memory on their own disk, with no cloud.
Good fit when
- Claude Code and Codex work on one project and the second agent must know the first one's decisions
- The agent remembers something wrong: you need to find the source conversation, correct the fact or mute it
- Valuable conversations sit in another tool's export or memory files and need importing
- Work on a confidential project calls for private mode for a while so nothing is kept
- The memory moves to another machine: the store transfers by copying one directory
Not a fit when
- You need a short project note: a `CLAUDE.md` or `AGENTS.md` file is enough
- You need cloud memory synced across devices: Scope Recall stores data locally and moves by copying
- There is no Python: installation goes through `pip` into the environment the hooks run from
- You need full fact extraction without configuring models: without model routes and a budget only original-text search works
Example request
Find in the shared memory why we chose SQLite for the queue and show the source conversation that fact came fromLimitations
Installation goes through `pip` into the Python environment the hooks will use; `scope-recall setup` shows a plan and asks for confirmation, and the host asks permission for hooks or the MCP server on first launch. Without model routes and a budget only original-text search works, and full memory features switch on through `--configure-models`. The description and part of the documentation are in Chinese. The repository tests check contracts and a synthetic benchmark, not extraction quality or behavior on your private conversations.
How to disable. Run `scope-recall plan-uninstall --instance-root <path>`, review the plan and apply it with `apply-uninstall`: by default only the plugin wrapper files go and the memory stays. Deleting the data takes a separate `plan-uninstall --purge` and `apply-uninstall --purge` pair.
FAQ
How do I install Scope Recall?
In the environment you want, run `python -m pip install -U "hermes-scope-recall[codex]"`, then `scope-recall setup` and confirm the plan. In a new host session approve the plugin hooks or the MCP server.
Does it work with Claude Code?
Yes, Claude Code attaches as a host: a separate guide sits in `docs/install/claude-code.md` and the shared store attaches with `attach --host claude-code`.
How do I delete a memory?
The `scope-recall memory` command finds facts, corrects and mutes them or deletes a conversation. Deletion first lists dependent sources and facts and applies only with `--yes`. `memory export --out` writes the visible store as a portable package.
Do my conversations go to external models?
Only if you configured model routes and a budget. Keys are environment references and setup never asks for a value. `private on --mode full` refuses capture, recall and background model work.
What about secrets that ended up in conversations?
`scope-recall memory secrets` lists credential matches by reference, provenance and time. Remove them with `scope-recall memory forget` and change any key found.
Discussion
No questions or reviews yet.
Related
PluginsLocal memory for agents with a review queue: SQLite, an audit log and a token ledger, one vault for Claude Code, Codex and Cursor
The session-recall command reads an agent’s local sessions and returns past conversations in compact form, with no MCP or hooks.
Local memory for Claude Code, Codex and Claude Desktop: one Markdown folder and one MCP server shared by all.
Coding memory across sessions, machines and teammates: decisions and dead ends flow into later prompts on their own.
A long-term memory plugin that keeps decisions, preferences and task state across sessions and injects only what a query needs
Agent memory and a knowledge graph in one SQLite file: the plugin adds /remember and /recall to Claude Code and Codex.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Scope Recall (this card) | no | 542 | ||
| Claude-Mem | no | 116,457 | a persistent-memory plugin for agent sessions, without one store shared with Codex and Hermes | |
| Basic Memory | no | not required | 4,566 | keeps memory in local Markdown files with a link graph |
| MCP Memory Service | no | not required | 2,117 | a self-hosted memory service with a knowledge graph |
README in short
The description of version 4.0.0 opens with a quick start and a host table, then sections on recall, management, private mode, agent-run migration and data boundaries. Details live in docs: per-host install, Docker, import, export and backups.