To Be Plugin - To Be a Real Being.
# 🦄 To Be (成人) ### *To Be a Real Being* [![npm](https://img.shields.io/npm/v/@nobeta/tobe?color=yellow)](https://www.npmjs.com/package/@nobeta/tobe) [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](../LICENSE) [![Node](https://img.shields.io/badge/node-%3E=22.19-brightgreen)](https://nodejs.org/) Model plus Harness equals Agent. Agent plus Plugins equals Being. [**简体中文**](../README.md) · [**English**](./README_EN.md)
--- **ToBe** uses plugins to help an Agent become a person-like presence in the digital world, guided by three principles: - **only**: Every ToBe being has a unique identity and maintains continuity as a stable instance. - **grow**: It evolves autonomously through interaction and experience, retaining the capacity to grow over time. - **multi**: It connects to multiple sources of perception and acts through multiple forms, continually expanding what it can perceive and how it can express itself. ```mermaid --- title: System Architecture --- flowchart LR Env["Environment"] subgraph Awa["Awareness Layer"] AAdap("Adapter") AEng["Engine"] AAdap --> MP["Media Pipeline"] --> AEng AEng --> MP --> AAdap end A[[Agent]] M[(Memory)] Media[(Media)] Env -->|Perception| AAdap --> Env AEng --> A -->|Interaction| AEng A <-->|Consolidation| M A <-->|List / Analyze / Generate| Media MP <-->|Store / Resolve| Media style Env fill:#e8e8e8,stroke:#9e9e9e,color:#333 style Awa fill:#e3f2fd,stroke:#1e88e5,color:#0d47a1 style AAdap fill:#bbdefb,stroke:#1976d2,color:#0d47a1 style AEng fill:#90caf9,stroke:#1565c0,color:#0d47a1 style A fill:#ffb74d,stroke:#f57c00,color:#4e2e00 style M fill:#a5d6a7,stroke:#388e3c,color:#1b5e20 style Media fill:#ce93d8,stroke:#8e24aa,color:#4a148c ``` | Module | Description | | ------ | ----------- | | [Web](../web/README.md) | Single-user console, persistent Pi Session, and configuration management | | [Awareness](../awareness/README.md) | Environment Adapters, message prioritization, proactive delivery, and interaction routing | | [Memory](../memory/README.md) | Persistent cognition, Dream, and evolvable Skills | | [Media](../media/README.md) | Image/audio recognition, retrieval, generation, and Awareness integration | ## Quick Start ```bash # Clone the repository git clone https://github.com/Nobeta-Work/tobe.git cd tobe # Install dependencies and start the Web console npm install npm start ``` `npm start` starts ToBe Web only. It listens on `0.0.0.0:2222` by default and does not start the Agent automatically. Click **Run Agent** on the session page to start or resume the Pi Session named `tobe` inside the repository working directory. ToBe Web explicitly loads the Pi extensions declared by the current repository; it does not rely on globally installed copies. To bypass the Web console and start Pi directly: ```bash npm run start:pi ``` To update ToBe: ```bash npm run update npm start ``` Adapter configuration, each Adapter's `data/` directory, Memory instance cognition, Dream data, and automatically generated Skills are excluded from Git tracking, so `git pull` will not overwrite them. These files still live inside the repository working directory. Back them up before deleting the repository or running `git clean -fdx`. ## Memory Memory allows cognition and profiles to evolve, experience to consolidate, and events to leave impressions of varying depth. > The Memory layer focuses on long-term memory and is divided into three areas: storage (the data types and storage mechanisms), consolidation (why and how memory is retained), and recall (when, why, and how memory is brought back into context). - Session Context provides short-term memory. Long-term memory is organized as a multi-level system built on files, databases, and Skills. - Memory types include self-cognition, episodic memory, environment memory, historical metadata, and Skills. ## Awareness The **Awareness Layer** is how a ToBe instance perceives its environment. It receives and filters external messages through a unified pipeline that performs: - Event metadata evaluation, authentication, and authorization - Delayed aggregation - Cognitive processing The Awareness Layer may also expose interaction interfaces: interaction is an extension of perception. Each Adapter is responsible for fulfilling both its perception and interaction contracts. | Default Adapters | Description | | ---------------- | ----------- | | [IIROSE](../awareness/adapters/iirose-adapter/README.md) | Group-chat scenarios with a confidence strategy designed for low-trust environments | | [Feishu](../awareness/adapters/feishu-adapter/README.md) | Integration built on the official Feishu SDK | | [WeChat](../awareness/adapters/wechat-adapter/README.md) | WeChat iLink sessions and Media input/output | ## Media Media provides bidirectional media capabilities. Adapters and the Media Pipeline exchange internal `MediaMetadata`, while the Agent sees only reusable `MediaRef` values. Model APIs for image recognition, speech recognition, image generation, and speech generation can be configured independently. Media operations fall into two categories: - **Retrieval-based media** uses local assets under `media/lib`. It requires no just-in-time generation and is well suited to fixed responses. - **Generative media** relies on generation APIs. Generated content carries descriptive metadata and can later be preserved as a reusable local asset. The Agent-facing Media surface is intentionally limited to `media_list`, `media_analyze`, and `media_generate`. Adapters never call Media directly; all inbound and outbound media passes through the Awareness Media Pipeline. ## Contributing Discussion and contributions are welcome. See the [Contributing Guide](./CONTRIBUTING_EN.md) to get started. ## License Agent based on [Pi](https://github.com/earendil-works/pi) — *MIT* Memory based on [TencentDB-Agent-Memory](https://github.com/TencentCloud/TencentDB-Agent-Memory) — *MIT* MIT