AI-powered asset intelligence for codebases.
Audit image assets across projects: find duplicates, unused files, oversized images, OCR text, AI tags, and safe next actions — locally and visually.

Install • Run UI • AI Canvas Demo • Why Aisets • AI • Roadmap

Important

Aisets is in active development. Expect fast iteration and review the release notes before updating.

AI Canvas Demo

AI Canvas turns asset review into a shared visual workspace, not just 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 are easier to trust.

Watch the AI Canvas toy-shop staging demo

Install

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

Quick Start

# 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.

Sharing & Remote Access

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:

# Remove the installed binary
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.

Why Aisets

Images look harmless until they become product debt.

They get copied between repos, renamed without context, exported at the wrong size, checked in “just for now”, referenced from code no one remembers, and kept forever because deleting them feels unsafe. After a few releases, your asset folder is no longer a folder — it is an undocumented dependency graph.

Aisets exists to make that graph visible and actionable:

• Know what you have — scan every project and see size, format, dimensions, usage, duplicates, optimization hints, OCR text, AI tags, and source context in one place.
• Find images the way humans remember them — search for “login screen”, “receipt photo”, “old hero banner”, or text inside screenshots even when filenames say nothing.
• Turn unknown files into labeled inventory — use AI to describe images, extract text, group themes, translate labels, and build semantic search data.
• Make cleanup reviewable — inspect duplicate groups, unused candidates, rename plans, and optimization opportunities before touching the filesystem.
• Use AI Canvas for visual decisions — gather related assets onto a board, compare variants, mark evidence, stage before/after cleanup plans, and capture a reviewable explanation.
• Reduce bloat with evidence — prioritize oversized rasters, heavy GIFs, missing responsive variants, lazy-loading issues, and high-impact format conversions.
• Handle real codebases, not toy folders — work across apps, design systems, asset packs, marketing sites, and long-lived monorepos.
• Stay local-first — run the dashboard on your machine and choose whether AI runs locally, through installed agent CLIs, or through compatible providers.

If your team has ever asked “can we delete this image?” and the honest answer was “I don’t know”, Aisets is built for that moment.

AI That Actually Helps

Aisets is not just another file browser with an AI badge. AI is used where it makes asset maintenance faster:

• 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 — search assets by intent, visual meaning, tags, descriptions, and OCR text.
• AI cleanup planning — use agent CLIs to help turn scan findings into practical cleanup plans.
• Content-aware workflows — build filters around AI category, OCR content, tags, duplicate status, and optimization potential.

Supported AI paths include:

• Local AI — connect local / OpenAI-compatible runtimes such as Ollama or LM Studio for private workflows.
• Agent CLIs — integrate with coding agents including Codex CLI and Pi, plus Claude Code, Cursor Agent, Gemini CLI, and Copilot CLI.
• Your choice — keep everything local, use agent tools already installed on your machine, or connect compatible providers when you want them.

Product Tour

Track workspace health	Search by meaning, not filenames

Turn metadata into reusable workflows	Audit tags and categories

Compare similar images side by side	Estimate optimization work

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

What You Can Do With It

Find duplicates before they become permanent

• Detect exact duplicate files across projects.
• Compare visually similar images side by side.
• Pick what to keep, preview the merge, and update references safely.

Delete with confidence

• Find assets that code no longer references.
• Separate code projects from asset packs and libraries so “unused” does not become reckless.
• Preview every destructive action before applying it.

Make images smaller without guessing

• Estimate savings before converting anything.
• Find oversized rasters, heavy GIFs, large inline imports, missing dimensions, and lazy-loading issues.
• Generate practical optimization scripts when you are ready.

Turn image chaos into searchable knowledge

• OCR screenshots and design exports.
• Tag images automatically with AI.
• Save filters for cleanup candidates, huge files, OCR matches, AI categories, duplicated assets, and review queues.

Check files before adding them

• Drop in new images before import.
• Catch duplicates, naming issues, and optimization opportunities early.
• Avoid adding asset debt in the first place.

What We're Building Next

Aisets already has scanning, AI metadata, OCR, embeddings, semantic search, safe actions, Image Tools, and an AI Canvas foundation. The next work is about turning those pieces into a more trustworthy AI asset workflow:

• AI operation safety contract — every AI-originated file or metadata change should be typed, validated by the backend, previewable, and auditable.
• Embedding index health — semantic search should clearly show whether text/image vectors are ready, stale, missing, or generated by a different provider/model.
• Assistant asset operations — natural-language requests should return grounded asset cards, comparison previews, cleanup plans, and safe operation cards instead of text-only answers.
• AI Canvas hardening — keep the visual workspace focused on asset work: arrange images, inspect context, create variants, annotate regions, and review operations before applying anything.
• Agent cleanup plans — use Codex CLI, Pi, and other agent CLIs to generate structured cleanup plans with preview commands, blockers, verification checks, and Aisets-side validation.

Safety First

Aisets is designed for repos you care about:

• File-changing actions use preview → confirm → apply.
• Apply rechecks files before writing, so stale previews are rejected.
• Project type controls how aggressive unused detection should be.
• Local-first workflows keep your assets on your machine unless you choose an external AI provider or agent.

License

MIT © Willie