SmartMemory for Obsidian

Bring SmartMemory's entity extraction, knowledge graph, multi-hop search, and contradiction detection into your Obsidian vault.

Requires a SmartMemory account. A free tier is available (1,000 notes, 200 searches/day). Paid tiers unlock higher quotas and team workspaces.

Features

• Vault ingest — turn any note into a structured memory item with extracted entities and relations.
• Search & Recall — multi-hop semantic search across your whole vault from a sidebar pane or Cmd+Shift+R floating modal.
• Entity backlinks — auto-extracted entities become clickable chips that find every note referencing the same person, project, or concept.
• Knowledge graph pane — focused N-hop neighborhood graph centered on the active note.
• Auto-link — propose [[wikilinks]] for entity mentions; preview-and-accept never modifies code blocks, frontmatter, or existing links.
• Inline suggestions — debounced "Related: Note A, Note B" widget while you type (opt-in).
• Contradiction banner — when a memory has been superseded server-side, you see it inline with a "View derivation history" button.

Install

Via BRAT (recommended while we are pre-directory)

1. Install the BRAT plugin from the community directory.
2. Open BRAT settings → "Add Beta Plugin" → paste smartmemory/smart-memory-obsidian → Add.
3. Enable "SmartMemory" in Settings → Community plugins.

Manual

1. Download main.js, manifest.json, and styles.css from the latest release.
2. Drop them into <your-vault>/.obsidian/plugins/smartmemory/.
3. Enable "SmartMemory" in Settings → Community plugins.

Coming soon

Direct install from Obsidian's Community plugin directory (submission pending).

Setup

1. Create a SmartMemory account at www.smartmemory.ai and copy your API key from Settings → API Keys.
2. In Obsidian: Settings → SmartMemory → paste API key, set workspace ID, save.
3. Status-bar dot turns green once connected.

API key can also be supplied via the SMARTMEMORY_API_KEY environment variable, which takes precedence over the settings field.

Privacy & data

• All network calls go to your configured SmartMemory API URL (default: https://api.smartmemory.ai).
• Note content sent on ingest, search, and recall is processed server-side for entity extraction and embedding. Your data lives in your SmartMemory workspace under your account.
• The plugin stores a local mapping (<plugin-data>/data.json) of vault path → SmartMemory item ID and a content hash for re-ingest detection. No vault content is logged anywhere outside Obsidian and your SmartMemory account.
• The plugin does not collect telemetry. The status-bar connection probe (a single GET /memory/list?limit=1 on load) is the only background call.

Quotas & free tier

• Free tier: 1,000 notes, 200 searches/day.
• Hitting the note cap: ingest returns 429; the plugin shows an upgrade modal.
• Hitting the daily search cap: search returns 429; a non-blocking notice surfaces.
• Both errors are non-destructive — the plugin never deletes vault content.

Settings reference

Setting	Purpose
API URL / API Key / Workspace ID	Connection
Auto-ingest on save / on create	Toggle automatic ingest of vault changes
Exclude folders	Comma-separated path prefixes (e.g. templates/, .obsidian/). Applies to all note sends and edits, including automatic ingest, recall, suggestions, and startup migration. Case-insensitive, includes subfolders, and accepts either slash style. An entry without a trailing / also matches sibling names that start with it (Private matches PrivateArchive/). Add / to match only that folder and its subfolders.
Write entities / relations / memory_type / sync timestamp	Per-field toggles for what gets written back to frontmatter
Suggestion frequency (off / 5s / 2s / 1s)	Inline-suggestion debounce
Contradiction banner	Toggle the supersession warning + sweep interval
Graph max nodes / max hops	Knowledge-graph pane bounds

Empty or frontmatter-only notes are skipped by automatic ingest. Ingest current note explains that the note is empty. Folder summaries count empty notes as skipped.

Compatibility

• Obsidian 1.5.0 and above.
• Desktop and mobile (the plugin sets isDesktopOnly: false; mobile is supported but the graph pane is bandwidth-heavier than the rest).

Build from source

Use Node.js 22 or newer and npm. The default layout has this plugin, smart-memory-sdk-js, and smart-memory-graph as sibling git checkouts. Install dependencies in the SDK and graph first, then in this plugin:

(cd ../smart-memory-sdk-js && npm install --ignore-scripts --no-package-lock)
(cd ../smart-memory-graph && npm install --ignore-scripts --no-package-lock)
npm install --ignore-scripts --no-package-lock
npm run build

For an isolated plugin worktree, set SMARTMEMORY_SDK_PATH to the absolute SDK checkout path. npm's local dependency installation also needs explicit paths when the sibling checkouts are elsewhere:

export SMARTMEMORY_SDK_PATH=/absolute/path/to/smart-memory-sdk-js
export SMARTMEMORY_BUILD_SCRATCH=/absolute/path/to/task/scratch
npm install --ignore-scripts --no-package-lock --no-save \
  "smartmemory-sdk-js@$SMARTMEMORY_SDK_PATH" \
  "@smartmemory/graph@/absolute/path/to/smart-memory-graph"
npm run build

Every production build copies the SDK's tracked files and non-ignored untracked files, plus its installed dependencies, to a disposable directory under SMARTMEMORY_BUILD_SCRATCH (default: .build/ in this plugin). It runs the SDK's npm run build there before bundling. The input SDK checkout and its dist/ remain untouched. Existing SDK dist/ is never copied or used. Both legacy smartmemory-sdk-js imports and scoped @smartmemory/sdk-js imports use the fresh build. The disposable copy is removed after success or failure.

Missing SDK checkout, git metadata, installed SDK dependencies, failed SDK compilation, or missing SDK dist exports stop the build with a diagnostic. The previous main.js is removed when a production build starts, so a failed build cannot leave an older bundle to publish.

Before preparing a release, run the build and check its first line:

npm run build
head -n 1 main.js
# // SmartMemory SDK git=<full commit SHA> dirty=false

The header records the SDK commit and whether tracked edits or non-ignored untracked files were present. Use a clean SDK checkout (dirty=false) for release provenance. Retain this header in the release's main.js, alongside manifest.json and styles.css. Building does not change the plugin version or publish a release. Dependency versions and the graph checkout still need to be held constant for byte-identical builds.

Run the focused build regression with the same environment variables:

npm test -- tests/sdk-build.test.ts

It uses the real SDK/compiler in scratch copies to check stale dist replacement, clean and dirty provenance, missing checkout, and SDK compilation failure. npm run dev retains the existing local dependency watch workflow. Production SDK rebuilding and provenance apply to npm run build.

Issues & feedback

github.com/smartmemory/smart-memory-obsidian/issues

License

MIT — see LICENSE.