// SOLUTIONS.GUKI

Multi-model AI workbench
GUKi

Hand one job to a team of models get one finished draft back

Split one task across models that fit, merge a usable draft, and let AI edit local files or servers under your approval.

DesktopWindowsPublic previewData stays local

Download for Windows See the UI

Windows 10/11 x64 · ~170 MB · in-app auto-update

// PRODUCT.DEMO

terminal — guki

$ guki run "full launch plan for a jewelry storefront"
→ split into 3 subtasks
→ 3 models in parallel (copy / visual / review)
✓ merged draft ready to use

◆ Multi-model collab◆ Agent edits files & servers◆ Local-first◆ Keys stay off our cloud◆ One-click undo◆ Multi-window tasks◆ Knowledge graph◆ Multi-model collab◆ Agent edits files & servers◆ Local-first◆ Keys stay off our cloud◆ One-click undo◆ Multi-window tasks◆ Knowledge graph

// BEFORE.AFTER

Before vs after

One model, endless follow-ups

Models split work, one merged draft

Copy-paste across AI windows

One workbench

Agent edits with no undo

Undo this round in one click

Keys sent to unknown services

Keys local; hidden from models

// CAPABILITIES

Six real capabilities

// 01

Multi-model collab

Split work across models, merge a draft; capsule picker by family, search, multi-select; “let GUKi pick a set”.

// 02

Agent that acts

Read/write local folders, SSH servers; plan first, you approve; self-heal before asking; pause if you’re away, resume later.

// 03

You stay in control

Three permission levels; servers can be read-only / confirm / deny; undo each round.

// 04

Local-first & key safety

Chats stay on your machine; keys in the OS credential store; secrets in configs become placeholders before the model sees them.

// 05

Memory that grows

Project memory, knowledge base, lessons learned; a graph that only draws real links.

// 06

Code & web, live

Streaming layout, code cards, live web preview (desktop/tablet/phone), console errors → “fix with AI”.

// SCREENS

Interface

Home: projects, chats, neutral starter cards

Model capsule: family groups, search, multi-select

// WORKFLOW

Three steps

Say the job

Describe the goal; pick models or let GUKi choose.

Models work in parallel

Every step visible: plan, tools, streaming output.

Take the draft

Undo or continue; file edits land in a change card.

// SECURITY.TS

Trust & control

▹Three permission levels
▹Windows sandbox for the backend
▹API keys in the OS credential store
▹Secrets redacted before model calls
▹In-app updates are separately signed

Has limits: passwords without key names may slip through. Missing EV/OV ≠ backdoor—it’s a signing-cost choice; this FAQ will change once signed.

// WHO.IT.FITS

Who it’s for

Cross-border store ops

Launch plans, competitor research, landing copy in one pass.

Developers

Read/edit code, preview pages, fix console errors with AI.

Content & research

Parallel drafts from several models into one usable piece.

Server ops

SSH in; approve the plan; optional read-only mode.

// COMPARE.TABLE

How it compares

Capability	GUKi	Single chat window	Manual multi-AI
Collab	Parallel + merge	Single-thread	You copy-paste
Agency	Local files / SSH	Mostly text	Varies by product
Undo	One-click round undo	No file-level undo	You back up yourself
Data location	Local-first	Often cloud	Scattered
Keys	OS store + redaction	Product-dependent	Easy to leak

// FAQ

FAQ

Chats, project memory, knowledge base, and run logs stay on your Windows PC by default. We do not run a “store your chats on GULINGLE cloud” service.

// DELIVERY.CARD

Version 0.1.0-beta.26
Platform Windows 10/11 x64
Form: desktop app
License: no paywall yet
Updates: in-app (signature-checked)

Multi-model AI workbenchGUKi