Scope Recall
Give supported coding assistants shared, local long-term memory with traceable sources and owner controls.
- Source repo
- 410979729/scope-recall-hermes
- Stars
- ★ 534
- Last updated
- 1d ago
- License
- MIT
- Primary language
- Python
- FA score
- 82/100 · Good
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code
- Cost
- Free, no paid service needed
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- Developers using both Codex and Claude Code who want session memories available from one SQLite store.
- Not a fit if
- Teams expecting semantic retrieval and background model processing without configuring model routes
- Users who need memories from separate projects, branches, or scopes automatically merged
- Source review
- 82/100 · Good
What does this agent do, and when should you use it?
Scope Recall is a local memory system shared by Claude Code, Codex, Hermes, WorkBuddy, and DeepSeek Harness. SQLite stores original conversations, while extracted facts and vectors remain traceable and rebuildable. Users install, configure, and manage it through the scope-recall CLI; it also exposes Hermes tools, a Codex MCP interface, and a local browser management page. Retrieval can combine lexical and vector channels, and automatic recall injects items that pass its thresholds; original-text search remains available without model routes or budget. Data lives in a verified Core data directory, with vector indexes kept as rebuildable companions. It fits developers who want to share, inspect, and maintain local memory across supported assistants, with model processing, identity binding, and host approval configured as needed.
Install hermes-scope-recall[codex] in the Python environment used by the hooks, then run scope-recall setup to review detected hosts, storage, scopes, model budget, and the worker plan before confirming. Once a host approves its hooks or MCP server, the system captures conversation text in SQLite; a background worker can extract facts, maintain vectors, and process indexing work. Session retrieval selects visible memories that pass lexical coverage or vector-similarity thresholds. Users can run scope-recall memory to browse, search, inspect history, correct, mute, or delete content. Hermes tools and Codex MCP expose read-only profile, entity, and trace views; the local UI supports browsing, reviewed actions, private rules, export, and backup status.
- Developers using both Codex and Claude Code who want session memories available from one SQLite store.
- Individuals who need to trace a remembered fact to a conversation, correct it, or delete related source material.
- Teams that want to start with original-text search before configuring model routes and budget for model features.
- Hermes or Codex users who need read-only
profile,entity, ortraceviews over admitted facts and relationships. - Users importing external memory files or conversation exports who want to preview their source and time range before writing to the local store.
How do you install or deploy this agent?
Install the Codex extra in the Python environment that will run the hooks, then run setup and review its plan. In a new host session, approve the plugin hooks or MCP server when prompted. The development checkout can be installed from the repository root; the package-index command installs a published release.
python -m pip install -U "hermes-scope-recall[codex]"
scope-recall setupFor a development checkout:
python -m pip install -U ".[codex]"To enable model features, use guided configuration or supply a routes file. Keys are environment references; setup does not ask for key values.
scope-recall setup --configure-models
scope-recall setup --models <route-file.json>How do you use this agent?
Start a new session in a supported host and approve hooks or MCP when prompted. Use the CLI to manage and inspect memory:
scope-recall memory
scope-recall ui
scope-recall private on --session <id> --mode fullExport the visible store:
scope-recall memory export --out <new directory or .zip>Use the read-only profile, entity, and trace views through Codex MCP or Hermes tools; the host integration handles the connection.
What are this agent's strengths and limitations?
- SQLite retains original conversations, and facts and vectors remain source-linked and rebuildable.
- Supports Claude Code, Codex, Hermes, WorkBuddy, and DeepSeek Harness, with multiple entries able to share one store.
- Original-text search works without model routes or budget, and the CLI supports local inspection, correction, and deletion.
- Private rules can cover a session, directory, or whole store, with full-refusal and read-only modes.
- Model routes and budget require separate configuration; without them, the full model-backed memory features are unavailable.
- Each host requires its corresponding hooks or MCP integration to be installed and approved, with identity and scopes bound.
- Changing the embedding model requires an explicit re-embedding run; opening a store does not authorize potentially paid reconstruction.
- The README documents multi-host integrations but provides no evidence of native ChatGPT, Claude AI, or named API integrations.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Scope Recall This agent | 82 · Good | CLIFree | ★ 534 | 1d ago | Python | Codex · Claude Code |
| MemOS 2.0 Stardust | 51 · Major gaps | Self-hosted serviceFreemium | ★ 12k | 1d ago | TypeScript | — |
| PowerContext | 82 · Good | CLIFree | ★ 1.5k | today | Python | Codex · Claude Code |
| Memoh | 50 · Major gaps | — | ★ 2.8k | today | Go | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
Installation asks the owner to review and confirm a plan; model budgets default to zero and credentials are environment references. Deletion previews dependents before --yes, uninstall retains memory by default, and purge requires a separate plan. The README describes source provenance and rebuild boundaries. Deductions: sensitive content is still kept in local long-term memory, and users are told to inspect screen text before indexing; dependency ranges are broad and the packet shows no dependency audit, so dependency security is only partially evidenced.
The README, CLI descriptions, and dsh plugin consistently describe bounded work, failure reporting, retry limits, and recoverable storage. Deduction: core storage and retrieval implementations are absent from this sample, so the stated guarantees cannot be checked across all paths; provider and host availability also depend on external systems.
The product names five host environments and covers shared stores, migration, private modes, original-text search, and optional model features with stated boundaries. Trigger guidance distinguishes automatic from explicit recall and specifies user-request requirements. Deduction: broad host support, Python versions, approval flows, and deployment modes require environment-specific setup and host reloads.
Repository layout, installation and host guides, configuration reference, management and migration documentation are substantial; MIT licensing and a security support/reporting policy are explicit. Deductions: the sample does not include changelog contents, maintainers are identified only as a contributor group, and comprehensive example/FAQ coverage cannot be established from the packet.
Source-backed recall is paired with inspection, revision, suppression, deletion, export, and profile/entity/path views, with meaningful boundaries described. Deductions: model features require route and budget setup, while SQLite, indexing, backups, and host integration add operating work; the described synthetic benchmark does not establish quality on private real-world conversations.
Many claims point to specific guides, configuration entries, or CLI behavior; the README separates what the synthetic benchmark can measure from semantic quality it cannot establish, and the memory skill requires tool-backed answers and evidence references. Deduction: key implementations and complete test results are outside this sample, so cross-source corroboration is partial and documentation claims are not treated as independent verification.
- This assessment uses only the supplied file sample and did not execute code; core storage and retrieval implementations are not included.
- Review screen-reading text before indexing. Deletion also removes the source message and facts derived from it, cannot be undone, and does not erase host history or existing backups.
- Before configuring model routes, check proxy settings, credential environment variables, and budgets; requests to external models may use the configured proxy.
Maintain this project? Add the FARS badge to your README
[](https://followagents.com/en/agents/scope-recall-hermes)<a href="https://followagents.com/en/agents/scope-recall-hermes"><img src="https://followagents.com/badge/scope-recall-hermes.svg" alt="FARS review"></a>Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
v4.0.3 scope-recall v4.0.3
Release notes excerpt
hermes memory setup and scope-recall setup --host hermes now give a new Hermes entry's memory to Hermes Desktop on the same machine as well as to the CLI: the entry gets the owner principal (desktop, local) and its owner_private row on the CLI row's scope, so a Desktop chat recalls and captures without a scope-recall grants edit. hermes --tui still needs the owner's approval, because the browser dashboard's chat tab runs the TUI and a dashboard on loopback can serve it without a login. Running setup again keeps an existing entry's grants; add Desktop to an entry made by an earlier …View on GitHub ↗ -
v4.0.2 scope-recall v4.0.2
Release notes excerpt
A Hermes that installs Scope Recall from its plugin catalog can now use it. hermes memory setup with scope-recall selected runs the ordinary setup through the provider's setup hook, using the plugin environment's interpreter, and reports a refusal in one line. Hermes entries in a shared store carry the Hermes profile's identity instead of a fixed agent id, so a fresh installation's sessions no longer fail with an agent_identity conflict; an unused entry recorded with the old id is repaired. Setup leaves a catalog-installed plugin directory under Hermes's control, treats a home that selec …View on GitHub ↗ -
v4.0.1 scope-recall v4.0.1
Release notes excerpt
Empty or whitespace-only HTTP 200 responses now get bounded recovery as unfinished upstream responses, while uncertain billing reservations and duplicate-execution safeguards remain in place. Hermes notices retain their trusted host origin across display timestamps and later human messages; those messages keep their own origin and are captured once. Scheduled Hermes sessions without a user no longer participate in memory and report capability_gap:unbound_session in diagnostics. After more than 60 seconds idle, binding a Hermes session warms the query embedding route in the background using t …View on GitHub ↗
FAQ
Can I use it without configuring a model provider?
Where is memory stored?
Can multiple assistants share one store?
Does importing external memory make it verified fact?
Thanks — we use this to improve the page.