Sinoclaw Agent: An AI Programming Assistant with Self-Learning Loop

Sinoclaw Agent is an AI programming agent based on Nous Research Hermes Agent. Its core innovation lies in the built-in self-learning loop, which can create skills from experience, persist knowledge across sessions, and search historical conversations. It supports multi-platform message gateways (Telegram, Discord, etc.) and multiple LLM backends, with flexible operating environments, making it an open-source, continuously learning AI programming partner.

Most current AI programming tools (e.g., GitHub Copilot) are 'single-inference' systems. Each conversation is isolated, unable to accumulate user preferences, project structures, or historical decision-making knowledge, so users have to repeatedly provide the same background information.

The core differentiation of Sinoclaw Agent is its self-learning loop, which includes: Agent Curated Memory (regularly reviewing and organizing knowledge), Autonomous Skill Creation (extracting reusable skills from complex tasks), Skill Self-Improvement (optimizing during use), FTS5 Conversation Search (cross-session recall), Honcho User Modeling (cross-session user profiling), and agentskills.io Compatibility (open skill standards).

• Flexible Deployment: Can run on $5 VPS, GPU clusters, or serverless infrastructure with low idle costs; supports terminal backends like local, Docker, SSH, and serverless persistence via Daytona/Modal.
• Multi-Model Support: Compatible with mainstream LLMs such as Nous Research Portal, OpenRouter (200+ models), and NVIDIA NIM, avoiding vendor lock-in.
• Message Gateway: Simultaneously connects to platforms like Telegram, Discord, Slack, enabling continuous cross-platform conversations and supporting voice transcription.

• Terminal TUI: Full terminal interface supporting multi-line editing, slash command completion, conversation history browsing, etc.
• Scheduled Automation: Built-in cron scheduler supporting natural language configuration of unattended tasks like daily reports and nightly backups.
• Parallelism and Delegation: Generates sub-agents to handle parallel workflows, and can call tools via Python script RPC to improve efficiency for complex tasks.

Sinoclaw Agent represents the evolution direction of AI programming assistants from 'intelligent completion' to 'continuously learning collaborators'. Its open-source nature, multi-model/cross-platform capabilities, and self-learning mechanism provide a customizable and extensible partner. It also includes research features like batch trajectory generation and Atropos RL environment, preparing data for the training of next-generation tool-calling models.

The project provides a one-click installation script that automatically handles dependencies like uv and Python 3.11; WSL2 is recommended for Windows users. After installation, start the CLI with hermes or the message gateway with sinoclaw gateway.