Noureddine RAMDI / Memex: a local-first AI-native knowledge management app with custom multi-agent orchestration

Created Mon, 04 May 2026 10:23:02 +0000 Modified Sat, 23 May 2026 20:41:27 +0000
1026 Words
github-stars flutter dart ai multi-agent local-first knowledge-management

memex-lab/memex

Memex is a local-first personal knowledge management app that combines AI-native features with multi-agent orchestration, all running on your phone without cloud dependencies. It captures text, photos, and voice inputs, then automatically organizes them into structured timeline cards enriched with extracted knowledge and insights. What makes Memex stand out is that it exposes the underlying agent infrastructure as a platform for users to create their own custom agents, using a combination of SKILL.md files, JavaScript, and event-driven triggers.

What memex does and its architecture

At its core, Memex is a Flutter app written in Dart, designed for both iOS and Android. It stores all data locally on the device using a combination of the filesystem and SQLite, ensuring that your data never leaves your phone unless you explicitly share it.

The app’s architecture centers around a multi-agent AI system. These agents operate as independent units that handle specific tasks such as knowledge extraction, card generation, and insight discovery. Each agent can be configured with different LLM models and providers, supporting more than 14 LLM providers including Gemini, OpenAI, Claude, Ollama, and several Chinese providers.

Memex’s agents communicate through event-driven triggers and can form dependency chains where one agent’s output feeds into another’s input. This design allows for complex workflows to be built and customized by users.

The multi-agent system is exposed to users as a platform. Custom agents can be defined via SKILL.md files, which follow Anthropic’s open Agent Skills standard. These files describe the agent’s behavior, system prompts, and interaction patterns. JavaScript support allows for dynamic execution as part of the agent’s operation, enabling flexible logic beyond static prompt templates.

This architecture allows Memex to be more than just a note-taking app — it acts as a programmable AI knowledge hub running fully on-device.

Why the multi-agent custom platform stands out

The most distinctive technical strength of Memex is its open multi-agent platform running locally. Unlike many AI-powered apps that treat AI as a black-box cloud service, Memex gives users direct access to the agent orchestration and customization layers.

This approach involves tradeoffs. Running all AI orchestration locally means the app depends on connecting to external LLM APIs, requiring API keys and internet access. However, it does not rely on any cloud storage or backend, preserving user privacy and control.

The codebase is surprisingly clean given the complexity of managing multiple agents, event triggers, and inter-agent dependencies. The use of Flutter/Dart means the app is cross-platform with a consistent UI and performance.

Supporting 14+ LLM providers with per-agent configuration adds complexity but also flexibility. Users can tailor each agent to a provider best suited for its task or region.

The SKILL.md files standardize how agents are described and allow for easy sharing and versioning of agent behaviors. This is a clever way to open up extensibility without requiring users to write full native code.

JavaScript execution within agents is a powerful feature that lets users implement conditional logic, data transformations, or custom interactions that go beyond static prompt templates. It effectively turns Memex into a programmable AI platform.

The tradeoff is that this flexibility comes with a steeper learning curve for users who want to create custom agents, and debugging multi-agent workflows can be complex.

Quick start

Memex provides several installation paths depending on your platform and preference:

  • iOS users can download it directly from the App Store (including a China-specific version).
  • Android users can get it from Google Play or download the latest APK from GitHub Releases for early access.
  • For developers, it can be built from source using Flutter.

To set up the development environment and build from source, you need Flutter SDK version 3.6.0 or higher, Xcode (for iOS), and Android Studio (for Android).

The basic setup commands are as follows:

git clone https://github.com/memex-lab/memex.git
cd memex
flutter pub get

For iOS, you also need to install CocoaPods dependencies:

cd ios && pod install && cd ..

After launching the app, you must configure your LLM API keys to enable AI features. This is done through the UI:

  1. Tap the avatar icon → Model Configuration
  2. Select your provider (Gemini, OpenAI, Claude, etc.)
  3. Enter your API key and optional base URL
  4. Configure each agent’s model independently

This per-agent model configuration lets you optimize which LLM powers which part of your knowledge workflow.

verdict

Memex is relevant for developers and power users who want a personal knowledge management system that is AI-powered, locally run, and highly customizable. Its multi-agent platform exposes advanced AI orchestration capabilities usually hidden behind proprietary cloud services.

The tradeoff is complexity: setting up and customizing agents requires learning the SKILL.md standard and some JavaScript, which might be overkill for casual note-takers.

The app’s local-first design respects user privacy and data control, a significant advantage for sensitive information. However, it still depends on external LLM APIs, so you need valid API keys and internet connectivity for AI features.

Overall, Memex offers a unique playground to experiment with multi-agent AI workflows on mobile devices, with a solid Flutter codebase and a thoughtful architecture that balances privacy, flexibility, and AI power. It’s worth exploring if you want more than a static note app and are comfortable tinkering with AI agents directly.

→ GitHub Repo: memex-lab/memex ⭐ 122 · Dart