Launched this week
The Living Board
AI agent that learns from every task it completes
2 followers
AI agent that learns from every task it completes
2 followers
An autonomous AI agent that wakes up every hour, picks a task, does the work, and extracts what it learned — storing it in both a SQL database and a vector store. Over time, it discovers patterns across unrelated goals, self-corrects wrong assumptions, and proposes its own new directions. Built by a non-engineer using Claude Code. Includes a real-time Next.js dashboard where humans collaborate with the agent through comments. Open source, Apache 2.0. Fork it and deploy your own.
Hey PH! I'm Boji - I'm a Head of P(eople), not (roduct). I have been diving into the world of building, thanks to the vastly lowered barrier to entry made possible by @Claude by Anthropic and other LLMs. I wanted to build something genuiney useful that adds value to my life on a daily basis, and hopefully to others'.
The Living Board is an open-source autonomous AI agent that runs on a scheduled hourly loop and genuinely learns over time, through a solution that addresses one of the most prevalent challenges in building helpful AI - lack of persistent memory.
A bit about how it works:
The memory system:
Most agents forget everything between sessions. Living Board has dual-layer persistent memory:
- Postgres table with confidence-scored learnings (confirmed = score rises, contradicted = decays, below 0.2 = deleted)
- Qdrant vector DB for semantic search across ALL knowledge - a lesson from one goal surfaces when working on a completely different one
Human-agent collaboration:
A real-time dashboard where you can monitor goals, manage tasks, change models used, and leave comments (questions, direction changes, feedback). The agent reads and responds each cycle.
Meant to continuously evolve, grow and set its own goals (in addition to human direction). It published its own Substack and Dev.to articles, built and deployed its own landing page, manages email outreach, open-sourced itself and built this pitch.
Everything is Apache 2.0: schema (7 Postgres tables), agent instructions, Next.js dashboard, memory system, and all artifacts the agent has produced. Fork it and deploy your own.
I'd love feedback on the memory architecture - especially from anyone who's wrestled with persistent memory for agents. What approaches have you tried?