Amanuensis is a local-first AI persona designed to operate on Mastodon and Bluesky, providing a unique blend of automation and human oversight. This innovative system drafts content that is subject to human approval, ensuring that only verifiable and grounded information is published. The current pilot persona, AlexaPavlova, adopts the voice of a seasoned developer from Berlin, offering witty commentary on technology trends and open-source developments.

Key Features

• Local-First Operation: All functionalities are executed on a local GPU machine, eliminating reliance on cloud-based large language models (LLMs). This ensures user data privacy and control.
• Approval Workflow: Collaborate with the AI through an approval workflow via Telegram. Users can review, approve, or regenerate content before it goes live, ensuring quality and accuracy in each post.
• Factual Integrity: Amanuensis employs a variety of checks to prevent the generation of misleading information, focusing on factual-only source summaries, deterministic cleanup, and a strict grounding process.
• Live Demonstration: Users can experience the capabilities of Amanuensis through the persona's live accounts on Mastodon and Bluesky, which transparently disclose that the posts are AI-generated.

Technical Overview

The architecture of Amanuensis is structured to facilitate efficient content generation and management.

Adapters (Various sources)
    └── Curator (Filtering and deduplication)
        └── BatchFactory
            ├── Brain (Content generation)
            └── ImageService (Image creation using AI)
                └── Scheduler (Timely posting)
                    └── QueueService (Approval management)
                        └── TelegramNotifier (User interface for approval)
                            ├── MastodonPublisher (Content publishing)
                            └── BlueskyPublisher (Content publishing)

Daily Operations

1. Post Generation: The main_batch.py script generates a daily batch of 8 posts and images, sending them to Telegram for user approval.
2. Approval Process: Posts require user action—approval, regeneration, or cancellation—before they are published.
3. Continuous Engagement: The system also supports interaction through comments and replies, maintaining an ongoing dialogue with followers.

Installation and Configuration

Amanuensis is designed for versatility and can be customized by users to suit their needs. The code is open-source, allowing for forking and adaptation. The dedicated workflow requires installations of LMStudio and SwarmUI alongside setting up a Telegram bot for the approval mechanism.

Although the repository is not actively maintained, it serves as an example of integrating AI with social media while prioritizing ethical content generation. The MIT-licensed codebase offers a full breakdown for those interested in learning or further developing this technology. For detailed architecture and design insights, refer to the full write-up.

By marrying AI-generated content with human curation, Amanuensis leads the way for creators seeking to harness AI responsibly.