Skip to content

kdeps desktop

Your AI agent,
on your desk.

The kdeps agent in a native window for macOS, Windows and Linux. Run a local model and nothing leaves your machine - or bring your own key. Drop in files, approve its actions, keep the history.

kdeps
lfm2.5gguf - local
Analyze: q3-report.pdf, budget.csv
read_file q3-report.pdf ok
load_document budget.csv ok
q3-report.pdf - 14-page quarterly report. Revenue up 8%, two cost lines over plan.
budget.csv - 212 rows; the same two lines are 11% and 6% over.
I can draft a variance summary or add the missing Q4 forecast column.
+

Message kdeps - drop files anywhere

Send

models

Every model kdeps knows

The picker lists what /model list shows. Local models download on first use, with progress in the chat.

156GGUF builds
159llamafiles
11cloud backends
llamafileGGUFOllamaOpenAI-compatibleAnthropicGoogleAWS BedrockCohereHugging FaceCloudflareIBM watsonxERNIEMaritacaMicrosoft 365 Copilot

why kdeps desktop

More than a chat window

It is the same agent loop the terminal runs, so tools, memory, sessions, goals and judges behave the same.

Drop a file, get an answer

Drag files or folders onto the window. The agent works out what each one is, reads it with the right tool - text, PDF, DOCX, CSV, code - and tells you what you can do with it.

An agent, not a chat box

The same loop as the terminal: shell, file edits, web search, SQL, memory, goals and a judge panel. Your workflows and agencies show up as tools it can call.

You approve every action

In ask mode each tool call and every path outside the workspace opens an approval: allow once, allow always, or deny.

Workspaces with memory

A workspace is a folder. History, search and persistent memory are kept per folder, and file tools never leave it.

Every setting, no YAML

Settings are generated from the config schema: dropdowns, sliders and suggestions for every field, plus harness sections, custom instructions and memories.

Take your setup anywhere

Export tuning, harness settings and skills as one profile file and import it on the next machine.

local vs hosted

Own the agent, not a subscription

What changes when the agent runs on your machine.

Hosted AI chatkdeps desktop
Your prompts and files leave the machineAlwaysNever, with a local model
Works offline / air-gappedNoYes - the UI loads nothing from a CDN
Edits files and runs commands for youIn a remote sandboxIn your workspace, with approval
Choice of modelThe vendor's300+ local builds, Ollama, 11 clouds
PriceSubscription or per tokenFree, Apache 2.0

getting started

Three steps

No CLI, no YAML, no account.

01

Install

Homebrew, Scoop, a Linux package, or the file from the release page.

->
02

Pick a model

Click the model name. Choose a local build to stay offline, or a cloud model and paste its key in Settings.

->
03

Ask or drop a file

Type a question, or drop files on the window and the agent starts analyzing them.

0requests to a CDN - the UI is embedded in the binary
3operating systems, one Go codebase
NoElectron - the system WebView
100%open source, Apache 2.0

download

Install kdeps desktop

Standalone - the kdeps CLI is not needed. Every release attaches the files with a .sha256 beside each.

macOS

Apple Silicon and Intel, macOS 11+

Download .dmg ->

Windows

x64, WebView2 (ships with Windows 11)

Download .zip ->

Linux

x86_64, GTK 3 + WebKitGTK 4.1

Download .deb / .rpm / .tar.gz ->

macOS builds are ad-hoc signed and Windows builds are unsigned, so the first launch needs right-click, Open (macOS) or More info, Run anyway (Windows). Full install and build guide ->

Stop renting your AI. Run it.

A local model, your files, your machine.

Released under the Apache 2.0 License.