×

Qwen3.8 Show and Tell

Sift

Find, understand, and organize your files with private local AI

By Richard Cornelius Suwandi

About me

Hi, I’m Richard

AI PhD studentCUHK-Shenzhen

Co-founderAIDDA Institute

Developer ambassadorQwen

I work on adaptive intelligence for decision making, optimization & discovery, and I enjoy building open-source projects

Websiterichardcsuwandi.github.io LinkedInlinkedin.com/in/richardcsuwandi Xx.com/richardcsuwandi

The problem

You remember what a file is, not what it is called

A folder that keeps growing

Downloads, screenshots, and exports pile up faster than anyone sorts them

A screenshot with no name

The filename tells you nothing about what is actually inside the image

A document you know exists

You remember the content, an invoice, a paper, a receipt, but not where it lives

Cloud AI can help, but it requires sending file contents and folder structure elsewhere

What is Sift

Search, understand, and organize in one local workflow

Sift runs local models over your files, combining semantic search with file operations

Local

Files and models stay on your Mac, powered by local AI

Reviewable

Every change is shown as a plan before it happens

Multimodal

Search covers documents, screenshots, and images

How I built it

Qwen3.8 Max as a pair programmer

Over a few weekends, I used Qwen across the whole codebase

Autonomous coding

Planned, implemented, and tested complete changes

Multi-step reasoning

Reasoned across the full pipeline, not just one file

Task-level collaboration

I described the outcome instead of every code edit

How it works

How Sift processes a request

You, or an agentAsk in plain English or call a command directly

→

SiftIndexes, searches, and drafts a plan with local AI

→

Your filesRead locally and changed only after approval

No upload, account, cloud call, or telemetry

Demo 1 of 4

Find what you remember

Describe what you remember; Sift returns matches with evidence

sift ~/Downloadslocal

Ready 269 files indexed

Demo 2 of 4

Search what is inside an image

A local vision model makes screenshots, receipts, and charts searchable

sift ~/Downloadslocal

Ready 269 files indexed, 41 images read

Demo 3 of 4

Organize with a sentence

One instruction becomes a plan for the whole folder

sift ~/Downloadslocal

Ready 269 files indexed

Demo 4 of 4

Review before anything changes

Approve, edit, postpone, or undo changes

sift ~/Downloadslocal

Ready 269 files indexed

Try it

Ask Sift something

Ask anything below

sift ~/Downloadslocal

Ready 269 files indexed

›

Under the hood

Three local models, three jobs

qwen3:4bAnswers questions and organizes folders

qwen3-embedding:0.6bTurns files into embeddings for meaning-based search

qwen2.5vl:3bReads text and content inside images

All three run through Ollama and can be swapped for another compatible local model with /model

Built for agents too

Every command also runs as a single call

Anything you can ask interactively also works as a plain command, so other tools and agents can call Sift directly

# ask a question, get an answer back
sift ask "Where is my March invoice?" --folder ~/Downloads

# draft a plan, save it as JSON, apply it later
sift plan "move all PDFs into Papers" --folder ~/Downloads --save papers.json
sift apply papers.json

# propose a full reorganization
sift organize ~/Downloads --save organize.json

An agent can draft the JSON plan, a person can review it, and Sift can apply it later

Why Sift

Compared with rule-based organizers

Area	Sift	Existing organizers
Search	Natural-language, with visible evidence	Filenames, metadata, rules
Organization	Proposes structure from the full batch	Predefined categories or conditions
Workflow	Plain-language requests, before and after preview	Menus, profiles, fixed rules
Privacy	Local AI models, files stay on your Mac	Cloud usage varies by product

Get started

Install and run Sift in a few commands

curl -fsSL https://ollama.com/install.sh | sh
open -a Ollama

git clone https://github.com/richardcsuwandi/sift.git
cd sift && uv tool install .

sift ~/Downloads

Prefer a GUI? Sift also ships as a native macOS app

What is next

Where Sift is headed

Windows and Linux

Extending beyond the initial macOS release

Connections and Explore

Related files and automatically discovered groups

MCP for agents

Let other agents use Sift’s local index as context

Thank you for listening!

Sift is free and open source

GitHubgithub.com/richardcsuwandi/sift Project pagerichardcsuwandi.github.io/sift