About Amber

Modern LLM interfaces are stateless by default. Every new conversation starts cold, and the current workarounds such as forcing huge context windows or flat RAG pipelines are token-expensive, hallucinatory, and horrible for privacy.

Amber is a desktop-native knowledge architecture that acts as a persistent, structured, and token-efficient memory layer for local and cloud-hosted LLMs.

Amber does not just store personal memory as a huge number of text files rather it organizes knowledge into a set of specialized Vaults. When a query is made, a set of Multi-Agent Collaborative Reinforcement Learning (MACRL) routing agents extracts the intended context and pulls only relevant information straying across domain boundaries with a set of Doors while removing outdated information.

The goal is not to give an AI a bigger context window, but to give it a better-shaped one.

Table of Contents

• Architecture Flow
• Getting Started
  • Prerequisites
  • Install
  • Run the desktop app (Tauri)
  • Run the UI only (Vite dev server)
  • Lint / typecheck
  • Rust checks (core)
• Before committing
• Community
• License

Architecture Flow

flowchart TD
    %% Custom Styling
    classDef user fill:#2d3436,stroke:#74b9ff,stroke-width:2px,color:#fff
    classDef agent fill:#6c5ce7,stroke:#a29bfe,stroke-width:2px,color:#fff
    classDef memory fill:#00b894,stroke:#55efc4,stroke-width:2px,color:#1e272e
    classDef privacy fill:#d63031,stroke:#ff7675,stroke-width:2px,color:#fff
    classDef inference fill:#0984e3,stroke:#74b9ff,stroke-width:2px,color:#fff

    %% 1. Input
    Canvas(["Unified Canvas (User Query)"]):::user
    
    %% 2. Intent & Routing
    Intent["Intent Classifier & MACRL Routing Agent"]:::agent
    Canvas --> Intent
    
    %% 3. Retrieval
    subgraph Storage [Amber Local Graph]
        direction TB
        Root[("Root Graph (Always Loaded)")]:::memory
        Vaults[("Scoped Vaults (e.g., Coding)")]:::memory
        Doors[("Cross-Vault Doors")]:::memory
        Root --- Vaults
        Vaults -.->|"Follows Linked Context"| Doors
    end
    
    Intent -->|"Proposes Scope"| Storage
    
    %% 4. Optimization & Privacy
    subgraph Assembly [Context Assembler]
        direction TB
        Decay["Decay Trimmer: Drops stale nodes to fit token budget"]:::agent
        Stub["Privacy Filter: Converts LOCKED nodes to Pointer Stubs"]:::privacy
    end
    
    Storage --> Decay
    Decay --> Stub
    
    %% 5. Handoff & Inference
    subgraph InferenceLayer [LLM Inference]
        direction TB
        Cloud["Cloud LLM (e.g., OpenAI/Anthropic)"]:::inference
        Local["Local LLM (e.g., Ollama)"]:::inference
    end
    
    Stub -->|"Injects Safe Context"| Cloud
    Stub -->|"Injects Full Context"| Local
    
    Cloud -->|"Returns output with {{STUBS}}"| Local
    Local -->|"Resolves stubs with local sensitive data"| Output(["Final Output to Canvas"]):::user
    
    %% 6. Continuous Memory Loop
    subgraph PostSession [Continuous Memory Loop]
        direction TB
        MemAgent["Memory Agent (Extracts & Dedupes)"]:::agent
        Diff["Memory Diff Panel (User Review)"]:::user
    end
    
    Output -->|"Background Update"| MemAgent
    MemAgent -->|"Proposes Changeset"| Diff
    Diff -->|"User Accepts -> New Snapshot"| Storage

Loading

Getting Started

Prerequisites

• Node.js: 24+
• Rust: stable toolchain (install via rustup)
• System deps (Linux only): you need WebKitGTK + a few build libs.

Example for Ubuntu/Debian:

sudo apt-get update
sudo apt-get install -y \
  libwebkit2gtk-4.1-dev build-essential curl wget file libxdo-dev libssl-dev \
  libayatana-appindicator3-dev librsvg2-dev patchelf

Install

From the repo root:

npm ci

Run the desktop app (Tauri)

npm run tauri dev

Run the UI only (Vite dev server)

npm run dev

Lint / typecheck

npm run lint
npx tsc --noEmit

Rust checks (core)

cd core
cargo fmt
cargo clippy
cargo test

Before committing

Amber uses a single cross-platform preflight gate that matches CI.

Windows (PowerShell) / macOS / Linux (bash)

# Auto-fix formatting first (recommended)
npm run preflight:fix

# Then commit
git add -A
git commit -m "your message"

If you want checks only (no auto-fixes), run:

npm run preflight

Community

Join our Discord Server to discuss features, get help with local LLM setups, report bugs, and chat with other Amber contributors!

License

This project is licensed under the AGPLv3 License - see the LICENSE file for details.