Skip to content

memi is the AI workbench for product designers.

Run Codex or Claude Code with project memory, design-system context, receipts, and Figma/FigJam handoff in one signed macOS app.

  • Codex
  • Claude
  • Figma
  • OpenCode

Open source Apple-signed by Humyn LLC Team Z4ZUZ884U3

npm latest
@memi-design/cli@1.1.0
Studio app
1.0.4
Homebrew cask
1.0.4
Included

One workbench for design-aware agent runs.

Open a project, choose an agent, watch the run spine, and turn useful output into editable design memory.

Run agents
Start Codex or Claude Code with the right setup.

Auth, permissions, models, and session controls stay visible.

codexclaudepermissions
Inspect the work
Read prompts, plans, tools, files, and cost as receipts.

The run spine stays compact until you need the raw log.

plantoolsfiles
Keep memory
Turn decisions, tokens, research, and notes into context.

Design memory is editable project state, not hidden prompt soup.

decisionstokensartifacts
Prepare boards
Export local Mermaid and FigJam-ready planning source.

External sync waits for approval; the source stays inspectable.

researchmermaidfigjam
Memory

AI design tools forget. memi keeps the product system readable.

Design memory

Research, specs, tokens, decisions, and reviews in one readable layer.

Markdown / YAML

Plain project memory that agents can read, diff, and reuse.

Codex + Claude

Primary agents with explicit auth, permissions, and receipts.

Research design

Findings become specs, context, and board-ready source.

PM Board

Local Mermaid and FigJam-ready source before external sync.

Figma bridge

Pull tokens, components, trees, and screenshots when needed.

Proof

A signed macOS app for agent runs and product memory.

Studio keeps the runtime, Codex and Claude sessions, design artifacts, Figma source, and run history in one desktop app.

Open design system
Skills

The polish lives in actual skill files.

memi carries taste as executable practice: capture the screen, score tenets, catch traps, and turn critique into the next tweak.

View skills
memi Studio screenshot used as UX audit evidence
score 84 clarity + feedback protected
memi ux audit studio.png --skill UX_TENETS_TRAPS
[observe] screenshot captured      [########--] 84%
[tenets] clarity feedback trust    [#######---] covered
[traps] missing state + copy drift  [fix ready]
[next] receipt, retry, tighter labels
Default SUPERPOWER.md
Autonomous design loop OBSERVE -> PLAN -> EXECUTE -> VALIDATE -> ITERATE
  • Code Connect first: get_code_connect_map — check existing component mappings before creating anything
  • 4. VALIDATE (Self-Healing — MANDATORY)
  • 1. Never skip self-healing — screenshot everything you create
Canvas FIGMA_AUDIT.md
Audit before handoff memi audit figma --tokens --accessibility
  • Audit a Figma file for design system consistency, accessibility compliance, token adoption, and Code Connect coverage. Produces actionable findings. Requires /figma-use.
  • 2. Component Consistency
  • Audit a Figma file for design system consistency, accessibility compliance, token adoption, and Code Connect coverage. Produces actionable findings. Requires /figma-use.
Critique UX_TENETS_TRAPS.md
Tenets and traps memi ux audit screenshot.png --json
  • Tenets are the qualities to protect:
  • Traps are the recurring failure modes to detect:
  • Use memi ux audit --json for a focused JSON report, or inspect the ux field in memi diagnose --json and memi fix plan --json.
Figma
Design lane

Figma stays contextual.

Connect the bridge when the run needs it. Keep local Mermaid and FigJam source visible before any external write.

Bridge port
9224
WebSocket server
Plugin
Active
memi Terminal
Events
174
since page load
Uptime
9:29
healthy
Port scanner (9223–9232)
9223 9224 · listening 9225 9226 9227 9228 9229 9230 9231 9232
Live event stream
  1. plugin-connectedIdentified memi Terminal
  2. tokens.pull171 tokens · 1.2s
  3. selection3 nodes inspected
  4. plugin-disconnectedreconnecting in 16.3s…
  5. plugin-connectedunknown via figma
Launch notes

Recent notes from the build.

Release hygiene, designer controls, and the product-design workbench model behind Studio.

All posts
Advanced
Advanced agents

Keep the power, hide the clutter.

Codex and Claude Code stay first-class. Ollama, Hermes, OpenCode, Cursor, and local tools stay available when the job calls for them.

Help build the Notes pack and the design-system memory layer.

Contribute on GitHub