Documentation

VisionLabs is a precision identity engine. Workspace is a chat with a swarm of expert AI design agents. Studio is the precision vector canvas where the work lives. Everything below — every tool, every shortcut, every model, every limit — documented in one place.

# What's new

VisionLabs ships new precision tools most weeks. Here's what's landed recently across Workspace and Studio — every one of these is live now.

# What is VisionLabs

Most design tools assume you start from a blank canvas. VisionLabs assumes identity work is construction — there is a grid, there are ratios, there is a documented system underneath every mark that lasts. The tools should reflect that.

VisionLabs is built around three surfaces. Workspace (at /v1) is where you talk to a swarm of expert AI design agents. Studio (at /v2) is the precision image & vector canvas where the result is refined, with real construction grids, bezier curves, and brand exports. Library (at /v3) is your design library plus a 2D/3D logo lab — saves, collections, moodboards, and the Lab mode for extruding logos in 3D with shaders. They flow into each other — generate a mark in Workspace, click Open in Studio, refine it, keep it in Library.

# The products

You move work between them with one click. Any image in Workspace has an Open in Studio button. Any export in Studio can be sent back to a Workspace conversation as context.

Library (at /v3) is the third surface — a local-first design library where everything you make is saved, organised into collections and moodboards, with palette extraction. Its Lab tab is a 2D/3D logo lab: extrude any SVG/PNG mark into 3D with chrome, glass, iridescent and holographic shader presets, lit and keyframable. Reach it from the sidebar in Workspace, or directly at /v3.

# Quickstart

1. Go to visionlabs.studio/v1 and create an account (email + password).
2. Verify your email (check inbox, click the link).
3. Open a new chat. Try a brief: "Construct a logo on a golden-ratio grid for a coffee roastery called Halfway."
4. When the swarm finishes, click any image and hit Open in Studio.
5. In Studio, the file is loaded with grids on. Refine, then ⌘+E to export.

# Core concepts

Conversation

A single chat thread inside Workspace. Holds your messages, the agents' responses, generated images, and uploaded files. Each conversation has its own brand memory (see Brand memory).

Swarm

The set of expert agents that respond to a brief together — Grid, Mark, Type, System, Critique, and Export agents. See The swarm.

Mode

A toggle that changes how the swarm behaves. Build generates. Plan proposes a structured plan without burning image credits. See Modes.

Project

A folder of related conversations for one client or initiative. Project context is shared across conversations.

Construction grid

A real geometric grid (golden ratio, baseline, isometric, dot, square) you can place under your work in Studio. Snapping respects whichever grid is on.

# Workspace overview

Workspace lives at /v1. It is a streaming AI chat with a sidebar of conversations on the left, the current chat in the centre, and a slide-out preview pane on the right when you click any generated image. The composer at the bottom supports text, images, PDFs, multi-file drop, voice dictation, attachments, and a queue (you can type your next prompt while the current one is still streaming).

Everything generated stays in the conversation forever. There is no separate assets tab — your chat history is your asset library. Use ⌘+K to jump between conversations.

# The swarm

Behind every prompt is a swarm of expert agents. Each one is a specialist. They work in parallel, critique each other, and merge their work before responding. You see one streamed answer — under the hood, six perspectives shaped it.

# Modes

Above the composer you'll find a mode toggle. Modes change how the swarm responds and what tools it can reach for.

Build mode (default)

Full toolset. Image generation is enabled. Banner generation, file exports, web search — all on. Use this when you want output.

Plan mode

Image generation and other expensive tools are stripped. The swarm proposes a structured plan instead of building. Cheaper, faster, useful when you want to think out loud or refine direction before committing image credits.

Code mode

Tells the swarm to lean into code output (HTML, CSS, SVG markup, React snippets). The preview pane renders live code previews. Great for landing pages and brand banners.

Plan & run, Act first, Ask first (behaviour)

Separately from the tool modes above, a chip in the composer controls how the swarm works through a task:

• Plan & run (the default) — the swarm lays out a short plan, then immediately carries out every step in the same turn. The plan renders as a live checklist that ticks off step-by-step as it works. Planning first measurably improves multi-step results.
• Act first — skip the plan and just generate right away.
• Ask first — propose a plan and wait for your approval before generating anything (best when a request would spend image credits and you want to confirm first).

# Models

You can pick the Claude model per conversation. Each tier maps to a sensible default.

# Image generation

Six image models are available through FAL.ai. The swarm picks the right one automatically based on your prompt, but you can override the choice from the model picker. Image generation starts on Pro — free accounts can chat, plan and research, but don't generate images or video.

• FLUX 1.1 Pro Ultra — photorealistic, high resolution.
• FLUX schnell — fast drafts.
• FLUX dev — balanced quality & speed.
• Recraft V4.1 — native SVG vector mode. Best for logos that go straight to Studio.
• Ideogram v3 — type rendering, posters.
• Stable Diffusion XL — quick experimentation.

Every generated image lands in the chat, openable in the preview pane (zoom, download, send to Studio).

# Banner generator

Ask the swarm for a brand banner ("a 1200×630 Instagram banner for our launch, dark palette, our wordmark centred") and it composes the layout from primitives, then renders a pixel-perfect PNG you can download. The output is repeatable and editable — you can tell the swarm to "make the headline 20% larger" and the new file regenerates in seconds.

# Built-in tools

The swarm can reach for any of these without leaving the chat:

• Web search — live Claude web search, useful for "what is this brand currently doing" research.
• Colour palette extractor — drop an image, get a 5-colour palette with hex + role labels.
• Colour palette builder — describe the mood, get a palette with accessibility-checked contrast.
• Font pairing — pick a serif, get three sans-serif pairings with rationale.
• Logo-grid SVG generator — drop a mark, get a construction-grid SVG overlay.
• Compare mode — pin two images side-by-side for review.
• Brand guidelines DOCX builder — full guidelines document from your conversation context.
• Hand-off ZIP pack — a single download with all assets, exports and tokens.
• Image edit — describe a change, the swarm applies it (in-painting, recolour, upscale).
• File generation — DOCX, PDF, SVG, JSON, plain text outputs.

# Slash commands

Type / in the composer to open a quick-action menu. Slash commands expand into rich, structured prompts so the swarm gets sharp instructions without you typing a sentence each time. Use the arrow keys to navigate, ↵ or Tab to confirm, Esc to close.

# Files & PDFs

Drop any PDF, image, or document onto the composer (or anywhere in the chat — there is a global drop overlay). Multiple files at once is supported. The swarm reads PDFs page-by-page, treats images as visual context, and stores everything in encrypted EU object storage tied to your account.

Attachments are private to your conversation. Collaborators on shared conversations can see them; nobody else can.

# Projects

Group related conversations into a project — typically one project per client or brand. Project-level memory means an agent in conversation B knows about decisions made in conversation A.

Create a project from the sidebar's + New project button. Move conversations in and out via the conversation menu.

# Brand memory

The swarm automatically pins decisions to memory whenever you confirm something — "we're going with the lime accent", "our primary mark is the geometric F", "no curves below 12pt". These facts persist in the conversation forever and are quoted back in subsequent turns.

Inspect or edit them via the brain icon at the top of the chat.

# Studio overview

Studio lives at /v2. It is a precision vector canvas with a left toolbar, a centre canvas, a layers panel on the right, and a properties strip at the top. Everything is keyboard-first. Pan with space, zoom with ⌘+scroll, drop in any SVG or PNG to start.

Files are autosaved to localStorage every few seconds — you can close the tab and pick up where you left off.

# Pen & vectors

The pen tool (P) draws bezier paths. Click for sharp corners, click-drag for smooth curves. Hold ⌥/Alt while dragging to break a handle's symmetry. Close a path by clicking the first anchor.

Boolean operations (union, subtract, intersect, exclude) are in the context menu when multiple paths are selected — they produce clean bezier output, not bitmaps.

Multi-select anchors with the direct-select tool (A) — drag a marquee, then move/scale/rotate the selection as a group.

# Construction grids

Studio ships with five real construction grids. Toggle them from the construction panel.

• Golden ratio — phi (1.618) rectangle divisions for classical proportional construction.
• Baseline — horizontal rhythm grid for type lock-up.
• Isometric — 30°/30°/30° three-axis grid for icons and mechanical illustration.
• Dot — quiet alignment cues, low visual noise.
• Square — classic pixel grid, snap units configurable.

# Snapping

Snapping is per-target — you can have anchor snapping on but edge snapping off, for example. Each target has a clear on/off chip in the snap panel.

• Grid — to whichever construction grid is active.
• Anchors — snap to other anchor points.
• Edges — snap to path edges.
• Centres — snap to bounding-box centres.
• Optical — smart alignment guides that detect optical (not mathematical) alignment.

The HUD shows live numeric feedback during any drag — dimensions, angle, snap target.

# Layers

Right-hand layer panel. Drag to reorder. Click the eye to hide, click the lock to lock. Group with ⌘+G, ungroup with ⌘+⇧+G.

# Templates

The template library lives at /v1/templates. Pick a starting point (badge logo, monogram, wordmark, abstract mark, isometric icon) and Studio opens with the grids and reference paths already in place.

# Brand export

Export with ⌘+E. Single artefact (SVG, PNG, WebP, JPG) or a full brand pack ZIP (mark on light + mark on dark + monochrome + favicon + apple-touch-icon + tokens JSON + README).

All SVG exports include a small invisible <metadata> watermark stamping the file as constructed in VisionLabs — useful for asset provenance, doesn't affect rendering.

# AI panel

The chat panel in Studio is canvas-aware — it sees what you've selected and can describe, critique, or generate suggestions for it. Ask "are the curve handles balanced?" and it inspects the actual bezier data. Ask "what would this look like with one more iteration of golden-ratio nesting?" and it sketches.

# Mobile

# Shared conversations

Any conversation can be shared with a collaborator. They get their own user account, their own quota, but full read or write access to the conversation you invited them to. Every assistant turn is attributed to whoever prompted it.

# Invites & roles

Click the share icon in the conversation header to invite. Two ways to invite:

• By email — Resend sends them a magic invite link. They create an account (or log in) and land directly in the conversation.
• By link — generates a one-time, expiring token. Useful for ad-hoc sharing.

Roles

Ownership is transferable. From the member list, click Transfer ownership next to any editor's name — the role swap happens immediately and is audit-logged.

# Presence & typing

When a collaborator opens a shared conversation, you'll see their avatar appear in the header within a second. As they type, a soft "… typing" indicator appears under the composer.

Each user has a deterministic grayscale tint — same user, same shade across reloads — used in author chips and on the side of their bubbles, so attribution is glanceable without being loud.

# "While you were away" digest

When you reopen a shared conversation after time away, a slim banner at the top tells you what changed: "3 new messages from Sara, 2 new images, 1 new file — since you last opened, 6 hours ago." Click it to jump to the first unread.

# Public share link

Need to send a single conversation to a non-account holder for review? Create a public read-only link from the conversation menu. It's a frozen snapshot — read, copy, but no editing or replies.

# Plans & pricing

Free is for chat, planning and research — image & video generation start on Pro. Messages count one per turn (a whole Plan & run still counts as one); images and videos draw from their own monthly quotas. See /v1/pricing for the full breakdown.

# Billing portal

Stripe-managed. From Settings → Billing → Manage subscription, you can update card, switch tier, view invoices, cancel. Cancellation keeps you on the paid tier until the period ends.

# Quotas

Quotas are per user, reset on the first of each month. Sharing a conversation does not pool quotas — each collaborator burns their own. If you hit a limit, the composer surfaces a modal with an upgrade CTA; pending messages aren't lost.

# PWA install

Both Workspace and Studio are installable as Progressive Web Apps. In Chrome/Edge: address bar → install icon. In Safari (iOS): Share → Add to Home Screen. Once installed, it runs in its own window, with offline shell + an app icon.

# Data export

We don't believe in vendor lock-in. From Settings → Privacy → Export my data, you get a full JSON dump of every conversation, message, attachment URL, and account fact in seconds. Re-import is on the roadmap; for now, the JSON is human- and machine-readable.

# Where your data lives

VisionLabs is built in and hosted in the EU. Your conversations, uploaded images, generated assets and brand facts live on EU infrastructure throughout the entire stack — database, object storage, edge hosting, transactional email.

LLM and image-model inference runs through vetted providers under data-processing agreements. Your data is never used to train any model. Generated images are processed by the inference vendor and stored on your account — they're yours, royalty-free, for commercial use.

A full list of sub-processors with their roles, regions, and DPAs is available on request at info@visionlabs.studio.

# GDPR

We are the Controller; sub-processors are Processors. Full DPA available on request at info@visionlabs.studio. Cookie banner is opt-in for non-essential cookies.

# Delete account

Settings → Account → Delete account. Hard delete, no soft state. Conversations, attachments, audit log entries, billing references — all purged within 30 days. Stripe customer record is anonymised, not deleted (legal requirement).

# Keyboard shortcuts

Workspace (v1)

Studio (v2)

# Glossary

• Anchor — a point on a vector path. Two handles control the bezier curvature.
• Brief — the prompt that kicks off a swarm response. Usually 1-3 sentences.
• Bezier — the mathematical curve underlying all vector paths in Studio.
• Mark — the logomark; the symbol part of a logo, independent of the wordmark.
• Optical alignment — alignment that looks right, even if mathematically off (e.g. circle nudged up because the eye sees the centre of mass, not the bounding box).
• Phi — the golden ratio constant, ~1.618.
• Token — a single design value (colour, spacing unit, font size) given a semantic name.

# Changelog

Public-facing release notes are posted on the blog. For the engineering log, see SPEC.md in the repo (shared with collaborators on request).