Proactive Digital Worker with Enhanced Memory and Formatting
Proactive Digital Worker with Enhanced Memory and Formatting
A digital worker tool that addresses the inefficiencies of current solutions by offering smart action retrieval, simple setup, proactive task management, and enhanced memory capabilities. The tool will also ensure consistent formatting of output artifacts, making it more user-friendly and efficient.
Key Features
- Smart action retrieval to save tokens
- Simple setup and usage with a browser interface
- Proactive task planning and initiation
- Enhanced memory capabilities to consolidate daily events
- Consistent formatting of output artifacts
Related Problems (1)
Inefficient Digital Worker Tools
Current digital worker tools like OpenClaw have several inefficiencies that hinder productivity and ease of use. Users often face high token usage, complex setup processes, and lack of proactive task management. Additionally, these tools often lack consistent formatting for output artifacts and have limited memory capabilities, leading to repeated mistakes and inefficiencies.Sources (1)
Our team spent more than 6 months building a proactive digital worker independently from OpenClaw (not an OpenClaw wrapper). It solves some of OpenClaw's current pain points: **Saves more tokens:** smart action retrieval and task mode switching mean the agent only loads what it needs, when it needs it. **Simple to set up and use:** clone the repo, run one command, and a browser interface launches. That's it. (Python and Node.js required though!) **Proactive:** a built-in planner that plans and initiates tasks for you, with your approval. Define your goals, and work toward them on a schedule. Different from heartbeat, this runs once per day to propose proactive tasks, while the heartbeat is used only for recurring tasks. **Memory:** every night 3AM, it "dreams" to consolidate the day's events into preferences, mistakes, events, and long-term memory about your tasks and habits. Here is the implementation in detail: store events in [EVENT.md](http://event.md/) \-> read the events and distill them in [MEMORY.md](http://memory.md/) \-> retrieve memory snippet with RAG -> agent reads the [MEMORY.md](http://memory.md/) if the snippets are relevant. [**FORMAT.md**](http://FORMAT.md)**:** Agent sometimes outputs artifacts (PDF, PowerPoint) without consistency. We integrated [FORMAT.md](http://format.md/) into CraftBot, so it can follow a format and design schema when generating files. etc, etc. However, is CraftBot better than OpenClaw? Likely not. Does CraftBot have parts that are better than OpenClaw? Sure. But for a very long period, we have been building toward a stable version and have not market our work publicly. The result is, 100 Github stars in 6 months + 650 commits, a small group of happy tester but no traction beyond that. So, I am gonna put this out today, you can check out how we built it, and give this proactive digital worker a try. Let us know your user experience, and I would love to build toward your preference. Happy automating! Repo link in the comment.