Mnemee

A design-forward visual workspace and cloud archive for digital creatives.

Who it's for

Designers, art directors, UI/UX engineers, and creative researchers who save visual references from across the web and need a single, high-fidelity repository to keep them organized and findable.

The problem

Creative tools are optimized for text. Designers save references via browser bookmarks (which break), social media saves (which get buried), and local desktop folders (which are a mess). There's no unified place that treats visual content as a first-class asset and preserves it permanently.

What it does

Mnemee ingests image and video uploads, direct video URLs, notes, and generic web links as durable visual assets. Provider-specific social capture and article extraction are future extensions of the implemented URL-resource pipeline. The Inbox renders the archive as a masonry grid, while collections provide an infinite spatial canvas for composing moodboards.

Core concepts

• Workspaces — Tenant-level containers. Personal vaults, studio spaces, etc. Completely isolated.
• Collections — Independent infinite canvases within a workspace. Each collection is a spatial moodboard such as "Acme Rebrand", "Typography Inspo", or "Spring Palette". This is the main organizational unit.
• Folders — First-class objects on a collection canvas that open nested canvases. Folder placement is independent from the placement of its contents.
• Assets — The polymorphic card that currently unifies image and video uploads, generic link bookmarks, and notes, with richer resource types added behind the same card boundary.

Ingest

• Image upload — Drag and drop into the Inbox or at a chosen collection canvas position. Stored in cloud storage, full resolution preserved.
• Video upload/import — Add a local MP4/WebM or a direct file URL. A poster anchors the board card; the original remains available to play and download.
• Link bookmark — Paste or drop an HTTP(S) URL. Mnemee creates a card immediately, resolves generic metadata in the background, and stores safe responsive preview variants. The original link remains usable if resolution fails.
• Provider and article ingestion (planned) — Rich social resolution, readable-content extraction, and indexing build on the same resource and resolver boundaries; they are not part of generic unfurling.
• Notes — Quick text snippets with color labels for raw ideas.

Surface & find

• Color search — Extracted image palettes provide ranked retrieval in the Inbox and current collection/folder canvas. It preserves authored canvas positions and Inbox ordering. See how it works.
• Smart collections — Saved queries (by color, type, tags) that auto-populate.
• Full-text search across article content and notes.

Architecture

Mnemee is multi-tenant from day one. Every asset, collection, and folder belongs to a workspace and has creator metadata. Collections contain a spatial tree of nodes: image, video, note, and link assets are leaf nodes, while folders organize nested content. Each placement has an authored position on its collection or folder canvas. Folders are not assets.

The application consists of a React/Vite client, a Bun/Hono API backed by Postgres and Drizzle, plus AWS Lambda workers that process S3 uploads and external URL resources asynchronously through independent SQS queues. Collection and folder badges show descendant asset counts: images, notes, and links count; folders do not. The client uses XYFlow for collection canvas rendering and interaction while Mnemee's API remains the source of truth for node identity, hierarchy, and position.

See the engineering docs for a codebase map, architecture, data model, backend conventions, operations, and contribution workflow.

The real-AWS development, deployment, stage, secret, and React/Vite hosting workflow is documented in SST deployment.

Development

From the repository root, choose the development mode you need:

bun run dev         # local Vite + personal hybrid SST Live backend
bun run dev:client  # local Vite + stable deployed cloud backend
bun run dev:website # local Astro marketing website

bun run dev stops both processes when you press Ctrl-C. Package commands are also available for isolated work:

bun install              # once, at the repository root: installs every package
cd apps/web && bun run dev
cd apps/server && bun run dev

For normal end-to-end development, use real AWS S3 and SQS through SST instead of the isolated package commands; see SST deployment.

Run every package's lint, typecheck, format and tests from the repository root with bun run check, which uses Turborepo and replays cached results for packages that haven't changed. Inside a package, bun run lint, bun run typecheck, bun run format and bun run test run them for that package alone. CI runs the same checks through Turbo, only for the packages a change touches and the packages that depend on them. Running bun install at the repository root installs every workspace package and the tracked pre-commit hook, which formats staged source files with Oxfmt.

Shared dev tools (TypeScript, Oxlint, Oxfmt, Vitest, @types/node, @types/aws-lambda) have one version, set in the catalog of the root package.json. Packages list them as "catalog:", so a version bump happens in one place.

The marketing site at mnemee.space lives in apps/website/ (Astro, deployed to Cloudflare Workers). Run it with bun run dev:website and deploy it with bun run --cwd apps/website deploy.