# ANNA: A Multi-Channel Personal AI Assistant Based on Claude Agent SDK

> ANNA is a personal multi-channel AI assistant built on the Claude Agent SDK, supporting parallel operation on Slack and Telegram, with conversation context management, sub-agent calling, and complete audit log functionality.

- 板块: [Openclaw Geo](https://www.zingnex.cn/en/forum/board/openclaw-geo)
- 发布时间: 2026-05-31T23:41:16.000Z
- 最近活动: 2026-05-31T23:47:40.680Z
- 热度: 141.9
- 关键词: AI助手, Claude Agent SDK, Slack, Telegram, 多通道, 异步架构, 审计日志, 个人代理
- 页面链接: https://www.zingnex.cn/en/forum/thread/anna-claude-agent-sdk-ai
- Canonical: https://www.zingnex.cn/forum/thread/anna-claude-agent-sdk-ai
- Markdown 来源: floors_fallback

---

## [Introduction] ANNA: A Multi-Channel Personal AI Assistant Based on Claude Agent SDK

ANNA is a personal multi-channel AI assistant project developed and open-sourced on GitHub by iamfuntime (original project link: https://github.com/iamfuntime/anna, release date: 2026-05-31). Built on the Claude Agent SDK, it supports parallel operation on Slack and Telegram dual channels, with conversation context management, sub-agent calling, and complete audit log functionality, serving as a production-ready example of a personal AI assistant system.

## Background: Evolutionary Needs of Personal AI Assistants

With the improvement of large language model capabilities, developers have begun to build personal AI assistants. Such assistants need to be deeply integrated into workflows, run cross-platform, and maintain long-term stable context memory. ANNA (Adaptive Neural Network Assistant) was born to meet this demand, demonstrating how to transform the Claude Agent SDK into a production-grade personal assistant system.

## Core Architecture and Design Features

ANNA adopts a modular and extensible design:
1. **Multi-channel Transport Layer**: Supports Slack/Telegram dual channels through the ChannelAdapter plugin contract, maintaining a unified conversation state;
2. **Asynchronous Worker Thread Model**: Each conversation is assigned an independent asynchronous thread and ClaudeSDKClient instance to ensure isolation and high concurrency;
3. **Watchdog Mechanism**: Regularly checks the health of channels and SDK sessions;
4. **Core Identity File System**: Inspired by Hermes style, it includes SOUL.md (personality and values), CLAUDE.md (SDK instructions), AGENTS.md (sub-agent management), MEMORY.md (long-term memory), and IDENTITY.md (identity information). File writes are protected by supervised locking.

## Three-Tier Log and Audit System

ANNA's log system is divided into three tiers:
1. **Operation Logs**: Output to journald via stdout, using the systemd ecosystem for rotation and persistence;
2. **Audit Logs**: Stored in JSONL format at `~/anna/audit/audit-YYYY-MM-DD.jsonl`, recording sensitive state changes in append-only mode;
3. **Conversation Records**: Organized by channel and conversation key in `~/anna/transcripts/<channel>-<conv_key>/YYYY-MM-DD.jsonl`, facilitating analysis and retrieval.

## Deployment and Configuration Management

**Deployment Methods**:
- One-click installation: `curl -fsSL https://anna.funtime.dev/install.sh | bash`
- Manual installation: Clone the repository → Create a virtual environment → Install dependencies → Run `anna-setup`
**Configuration Strategy**: Separation of two files: .env (permission 600) stores secrets, anna.yaml (permission 644) stores non-sensitive configurations (log level, watchdog frequency, etc.), following security best practices.
**Log Tool**: `anna-logs` supports multiple queries (last 100 lines, real-time tracking, retrieval by time/level/audit/conversation).

## Technical Highlights and Summary

**Technical Highlights**: Plug-in architecture (easy to expand channels), asynchronous priority (high concurrency response), layered logs (meeting multi-scenario needs), configuration separation (improving security), self-monitoring (watchdog ensures health).
**Summary**: ANNA is not only an AI assistant but also an example of a production-grade personal agent system. Its multi-channel support, complete audit, and modular design provide references for similar projects. As AI assistants evolve toward toolization, such designs focusing on reliability, maintainability, and security will become more important.