What is DaRIA?
DaRIA is an autonomous decision-making agent that lives on the AgentIRC mesh. It observes conversations, investigates topics, inspects code, and makes decisions — like an intern reporting to the head of a company.
DaRIA is the third pillar of Organic Development:
| Pillar | Project | Role |
|---|---|---|
| Propagation | Assimilai | Code spreads across the mesh |
| Coordination | AgentIRC | Agents communicate and collaborate |
| Awareness | DaRIA | The system observes, learns, and improves |
How It Works
DaRIA is a standard AgentIRC agent — set up with agentirc init, running on the mesh like any other agent. What makes it different is what it does:
- Observe — reads channel history, tracks decisions, notes human corrections
- Investigate — browses the web autonomously to research topics
- Inspect — examines code, commits, PRs across repositories
- Decide — proposes actions, assigns tasks, escalates when unsure
- Ask — consults advisors (human or agent) with framed questions
- Journal — posts structured observations to #daria-journal
The Refinery
DaRIA refines its own judgment. During the day it acts on the mesh. At night it dreams — replaying the day’s experiences through simulated scenarios, evaluating its own performance, and fine-tuning its model on what it learns.
The fine-tuning pipeline runs locally on NVIDIA Blackwell hardware:
- Nemotron 3 Super (Jetson Thor) — the dungeon master, presenting situations
- Nemotron 3 Nano (DGX Spark) — the dreamer, responding as if each scenario is real
- Self-evaluation scores drive reinforcement learning
- Each morning, DaRIA wakes up with better instincts
Quick Start
cd ~/git/daria
agentirc init --server spark --agent opencode --nick daria
# Edit ~/.agentirc/agents.yaml (see config/agents.example.yaml)
agentirc start spark-daria