memi

Product Introduction

1. Definition: memi is a native macOS application that serves as an AI-powered workbench and product design memory layer. It is designed to run advanced AI agents like Claude Code and Codex locally on the user's hardware, integrating with design specifications, research documents, and live Figma files.
2. Core Value Proposition: The core value proposition is to unify fragmented AI coding tools and preserve design system integrity by providing a single, secure desktop environment for AI-assisted product development. It solves the problem of "AI tools forgetting" by maintaining a persistent, editable, and human-readable project memory, turning ephemeral AI sessions into durable design context.

Main Features

1. Unified Agent Workbench: memi functions as a single command center for launching and managing AI coding agents. It supports primary agents like Claude Code and OpenAI Codex, with configurable authentication, granular permissions, and session controls for each run. The workbench materializes the runtime environment locally on the user's Mac, ensuring all agent activities—from prompts to file edits—are contained within a transparent, auditable workflow.
2. Run Spine & Receipts: This feature provides deep observability into AI agent operations. It displays a live "run spine" that tracks the agent's steps, including plans, tool calls, file changes, and associated costs as itemized receipts. Users can inspect the raw logs for full auditability, moving from a compact summary to detailed technical data when needed, ensuring complete transparency over autonomous AI actions.
3. Editable Design Memory (memi://protocol): Unlike hidden prompt histories, memi creates a structured, editable project memory layer. It captures decisions, research findings, design tokens, and technical artifacts in plain Markdown and YAML formats. This "design memory" is diffable, reusable by AI agents, and serves as the single source of truth for the product system, preventing context drift between design, engineering, and AI tools.

Problems Solved

1. Pain Point: Product designers and engineers struggle with context fragmentation and tool sprawl. AI coding assistants operate in isolation, losing project-specific design system knowledge, while design files (Figma) and implementation details live in separate silos, leading to inconsistent outputs and repeated manual context loading.
2. Target Audience: The primary users are product designers, UI/UX engineers, and technical product leaders who leverage AI coding assistants (like Claude or Codex) but require their outputs to adhere to strict, evolving design systems and product specifications.
3. Use Cases: Essential for maintaining design system consistency during AI-augmented feature development; auditing Figma files for accessibility compliance and token adoption; creating an inspectable audit trail for AI-driven code generation; and preparing Mermaid diagrams and FigJam-ready planning sources for team collaboration.

Unique Advantages

1. Differentiation: Unlike generic AI coding terminals or plugins, memi is a design-system-aware, macOS-native workbench. It provides a signed, secure application that manages the entire lifecycle of an AI run—from agent selection and permissioning to output archival—while keeping the design system as an active, integrated layer rather than a static reference document.
2. Key Innovation: The key innovation is the bidirectional integration of live Figma data with local AI agent sessions via a secure bridge (WebSocket on localhost). This allows agents to pull component trees, design tokens, and screenshots contextually during a run. Furthermore, the "Proof" feature enables proactive AI auditing of design files for consistency, accessibility, and token usage, turning the AI from a passive code generator into an active guardian of design quality.

Frequently Asked Questions (FAQ)

1. What is memi and how does it differ from a standard AI coding tool? memi is a dedicated macOS workbench that acts as a manager for AI agents like Claude and Codex. Unlike a standard terminal, it provides persistent, editable project memory in plain Markdown/YAML, integrated Figma context, and auditable "run receipts," ensuring AI work aligns with your product's design system.
2. How does memi integrate with Figma for design work? memi runs a local, secure WebSocket bridge (e.g., on port 9224). You can connect it on-demand to a Figma file to pull design tokens, component structures, and screenshots directly into an AI agent's context. All synchronization is explicit and requires approval, keeping your source files safe.
3. Is memi secure and open source? memi Studio is a signed macOS application (by Humyn LLC), ensuring it is distributed via a verified developer identity. The core CLI and components are open source, allowing for transparency and community contribution, with the runtime and agent sessions kept local on your machine.
4. What AI agents can I run inside memi? memi primarily supports Claude Code and OpenAI Codex as first-class agents, managing their authentication and sessions. It also provides infrastructure to connect other agents and local tools like Ollama (Hermes), OpenCode, or Cursor, making it a flexible platform for various AI workflows.