smartmem-dev
75adca7024
Found live by DEMO-WALKTHROUGH-4 spike 0.6 against the local full stack: - The service rejects POST /memory/ingest without X-Workspace-Id (team_required policy; the API key alone carries no team context), so every cloud sync failed with HTTP 400. New Workspace ID setting (options page, storage.sync), sent whenever configured, fail-fast required for non-loopback targets. - Captures were stamped origin 'browser:save' — unregistered in core origin_policy, falling to tier 4: clips that landed were invisible to both search and recall. Now 'capture:extension', the registered tier-1 namespace. - README/QUICKSTART document the Workspace ID requirement. Manifest 1.2.2.
Smart Memory Capture
Version: 1.2.1
Capture surfaces that send web and mobile content into SmartMemory.
Capture is a thin ingestion surface — it extracts content well at the edge and posts it
to SmartMemory's /memory/ingest. All processing (extraction, entities, embeddings, the
knowledge graph) and storage happen in SmartMemory core. There is no separate database
or pipeline in this repo; captures appear in SmartMemory recall, not a local clip list.
Components
| Path | What it is |
|---|---|
extension/ |
Chrome/Edge/Brave MV3 browser extension (the primary surface) |
mobile-android/ |
Android share-sheet app (Capacitor) |
How it connects (CAP-CONN-1)
Both surfaces speak one REST contract — POST /memory/ingest — implemented by either
target. The extension classifies the target by endpoint host and applies the matching rule:
| Target | Endpoint | Auth |
|---|---|---|
| Local (lite) | http://127.0.0.1:9014 — a SmartMemory daemon on this machine |
No token (loopback is OS-authenticated) |
| Cloud (service) | https://api.smartmemory.ai (or self-hosted, HTTPS) |
API token + Workspace ID required (sent as X-Workspace-Id — the service scopes every capture to a workspace) |
- Detect local SmartMemory probes
GET /healthon the lite daemon and verifies it identifies as SmartMemory. If found and no endpoint is configured, it auto-targets the daemon. If not found, the UI guides you to start it (smartmemory server) or use Cloud. - The same backend can run on a SQLite (lite) or FalkorDB backend — transparent to capture.
"Lite" is the local SmartMemory distribution (the
smartmemorypackage's daemon), not a separate clip store. The browser extension needs an HTTP endpoint, so a fully embedded install with no running daemon is reachable only via SmartMemory's CLI/MCP, not this extension.
Browser extension
Install (unpacked)
- Open
chrome://extensions/, enable Developer mode - Load unpacked → select the
extension/folder - Open the extension's Settings:
- Click Detect local SmartMemory to use a local daemon, or
- Enter your Cloud API URL + API token
Capture modes
- Full Page — page text content
- Article — main article via Readability (great for news/blogs)
- Selection — selected text (right-click → "Capture selection to Smart Memory")
- Screenshot — visible-area snapshot
Captures carry rich provenance into SmartMemory context/properties: origin=browser:save
(search-visible user content), source_url, title, capture_mode, plus author / OG / video
metadata where available. Re-capturing the same URL dedupes (CORE-INGEST-DEDUPE-1).
Settings
- Storage Mode — Local First (cache locally, sync in background) or Server Primary (sync first)
- Local SmartMemory — detect/auto-target the lite daemon
- API URL / API Token — Cloud target (token required) or a local daemon URL
- Default tags, Duplicate URL policy, Export/Import
Privacy
Captures go only to the SmartMemory endpoint you configure (your local daemon or your account). Nothing is sent to any third party. The API token is stored device-locally and is never synced to Google's cloud.
Documentation
Full SmartMemory documentation: https://docs.smartmemory.ai