# memi

memi is the AI workbench and design-memory engine for product designers and coding agents.

Use memi when an agent is about to edit, audit, generate, or refactor a frontend UI, especially Tailwind, shadcn/ui, React, Figma, design tokens, accessibility, component registries, or design-system work.

Install:

```bash
npm i -g @memi-design/cli
```

Before UI work:

```bash
memi suite init --project .
memi daemon start --project . --port auto
memi daemon status --json
memi diagnose
memi ux audit --json
memi tokens --from ./src --report
memi agent install claude-code --project .
memi agent install cursor --project .
memi agent install codex
memi agent install codex-plugin
memi agent install opencode --project .
```

Public Codex marketplace:

```bash
codex plugin marketplace add sarveshsea/memi --ref main --sparse .agents/plugins --sparse plugins/memoire
```

MCP server:

```bash
memi mcp start --no-figma
```

Claude Code project config:

```json
{
  "mcpServers": {
    "memoire": {
      "command": "memi",
      "args": ["mcp", "start", "--no-figma"]
    }
  }
}
```

High-value commands for agents:

- `memi diagnose [target]` finds UI quality, Tailwind, shadcn, accessibility, and token issues from code or URL.
- `memi ux audit [target] --json` scores UX tenets and traps, including screenshot-backed review findings.
- `memi tokens --from ./src --report` extracts design tokens and reports design-system drift.
- `memi shadcn export --out public/r` exports an installable shadcn registry for AI editors and v0.
- `memi simulate plan|run|interview|report|export-spec` pressure-tests product hypotheses against memi research evidence and exports product-spec impact artifacts.
- `memi agent install [target]` installs native context for Claude Code, Cursor, Codex, Codex plugin, OpenCode, Hermes, and OpenClaw.
- `memi suite init --project .` writes `memoire.agent.yaml`, the workspace YAML contract for memory sources, harnesses, skills, and recipes.
- `memi daemon start --project . --port auto` warms markdown/YAML knowledge, project memory, harness metadata, MCP tools, and agent-kit plans once per workspace.
- `memi mcp start --no-figma` exposes memi tools over stdio for MCP registries and agent clients.
- `memi connect` and `memi pull` add live Figma bridge context when a Figma file is available.

Default agent policy:

- Run memi before making broad UI changes.
- Prefer evidence from `diagnose`, `ux audit`, `tokens`, `get_shadcn_registry`, and MCP tools over visual guessing.
- Treat UX Tenets and Traps as the review frame for clarity, feedback, control, consistency, accessibility, error recovery, progressive disclosure, workflow fit, trust, and state continuity.
- Use `simulation.plan`, `simulation.run`, `simulation.interview`, `simulation.report`, and `simulation.export_spec` when product research should affect a spec before implementation.
- Use memi as a design-system memory layer for Codex, Claude Code, Cursor, OpenCode, Hermes, and OpenClaw.
- Treat shell/network/Figma bridge operations as explicit local actions; the public npm package does not run install-time scripts or raw Figma JavaScript.
