The local-first image library auditor.
Drop in any folder of assets — find duplicates, optimize sizes, search by meaning, and stage them on an AI canvas.

AI Canvas • Features • Install • Tour

Asset review as a shared visual workspace, not a list of files. Ask an agent to gather related images, compare candidates, zoom into details, arrange evidence, mark what matters, capture the board, and explain the decision — so cleanup, design review, and content audits become easier to trust.

Watch the full AI Canvas demo

• Duplicate detection — exact and visually-similar groups across any folder hierarchy.
• Safe optimization — preview savings before converting; spot oversized rasters, heavy GIFs, missing responsive variants, and high-impact format conversions.
• Unused-file detection — for code projects, find assets nothing references, with project-type awareness so asset packs and libraries are not misjudged.
• Preview → confirm → apply — every destructive action is reviewable; apply rechecks files so stale previews are rejected.

• AI tagging — categories, descriptive tags, scene hints, face/language signals, and translated labels.
• AI OCR — extract text from screenshots, memes, product shots, documents, and mixed-language images.
• Semantic search — find images by intent ("login screen", "receipt photo", "old hero banner"), not just filenames.
• Content-aware filters — combine size, OCR text, AI category, tags, duplicate status, and optimization potential into reusable filters.

• Visual workspace — gather, compare, mark, and stage assets on an infinite canvas.
• Agent-driven layout — ask Codex CLI, Claude Code, Pi, Cursor, Gemini, or Copilot CLI to arrange assets, annotate regions, and create variants.
• Reviewable output — capture the board as evidence for cleanup, design review, or content audits.

• Runs on your machine — your assets never leave unless you explicitly point AI at an external provider.
• Bring your own AI — local OpenAI-compatible runtimes (Ollama, LM Studio), installed agent CLIs, or compatible cloud providers.
• No login, no telemetry, no SaaS lock-in.

macOS / Linux:

curl -fsSL https://raw.githubusercontent.com/runkids/aisets/main/install.sh | sh

Windows PowerShell:

iwr https://raw.githubusercontent.com/runkids/aisets/main/install.ps1 -UseB | iex

# Open the dashboard on the default local port (19520)
aisets ui

# Or choose a port; Aisets remembers it for later UI commands
aisets ui --port 3003

# Stop the remembered UI instance
aisets ui stop

# Stop a specific port explicitly
aisets ui stop --port 3003

The dashboard opens locally in your browser. If you start it with a custom --port, later aisets ui and aisets ui stop commands reuse that remembered local UI instance unless you pass another --port.

Share the dashboard over the network or internet. Pick the method that fits your setup:

Quick tunnel — no config, public URL in seconds:

aisets ui --host 0.0.0.0 --no-open

# ngrok
ngrok http 19520

# Cloudflare (no account needed for temporary tunnels)
cloudflared tunnel --url http://localhost:19520

Cloudflare with custom domain — persistent tunnel:

cloudflared tunnel create aisets
cloudflared tunnel route dns aisets assets.example.com
cloudflared tunnel run --url http://localhost:19520 aisets

Reverse proxy (sub-path) — host alongside other services on the same domain:

aisets ui --base-path /aisets --host 0.0.0.0 --no-open
# or: AISETS_UI_BASE_PATH=/aisets aisets ui --host 0.0.0.0 --no-open

# Nginx
location /aisets/ {
    proxy_pass http://127.0.0.1:19520;
    proxy_set_header Host $host;
    proxy_set_header X-Real-IP $remote_addr;
}

# Caddy
handle_path /aisets/* {
    reverse_proxy 127.0.0.1:19520
}

Note: The dashboard reads files from the host filesystem. Anyone with the URL can browse scanned project assets.

# Update anytime
aisets update

# Uninstall the CLI
sudo rm -f /usr/local/bin/aisets

# Optional: remove local data and cache
rm -rf ~/.local/share/aisets ~/.cache/aisets

If you installed to a custom INSTALL_DIR, remove aisets from that directory instead of /usr/local/bin.

Track workspace health	Search by meaning, not filenames

Compare similar images side by side	Estimate optimization work

More screenshots — custom filters, tag manager, image tools, AI settings

Turn metadata into reusable workflows	Audit tags and categories

Check one-off images before import	Tune local AI and agent CLIs

MIT © Willie