About COSMO
COSMO is a task execution system with persistent knowledge. It's actively being developed.
What it is
COSMO stands for Cognitive Orchestration System for Metacognitive Operations. That's a mouthful, but it describes what the system does: orchestrate AI capabilities to execute structured work while maintaining awareness of what it's doing.
In practice, COSMO is:
- A plan-driven task execution system
- Built on two core agents (Research + IDE)
- Designed to produce persistent, queryable knowledge
- Focused on reliability and transparency
Current state
COSMO is in active development. Here's an honest assessment:
What works
- Plan-driven task execution with phase management
- Research Agent for web search and information gathering
- IDE Agent for file operations, analysis, and document creation
- Knowledge brain building with queryable results
- Real-time execution monitoring via WebSocket
- State management with retry logic and crash recovery
What's in progress
- Multi-user support and authentication
- Easier deployment and configuration
- Better user interface
- Documentation and onboarding
Recent focus
Recent development has been about reliability—making task execution more robust. The "IDE-first paradigm" simplifies the agent architecture from many specialists to two capable agents. This reduces complexity and makes the system more predictable.
Architecture evolution
COSMO's architecture has evolved. Earlier versions had many specialized agents: research, analysis, synthesis, document creation, code creation, and more. Each agent had narrow expertise.
The current direction consolidates to two agents:
- Research Agent — Handles external information (web search)
- IDE Agent — Handles everything else (files, analysis, synthesis, code, docs)
This "IDE-first" approach trades specialization for simplicity. One capable agent doing most work is easier to coordinate than many narrow specialists. The system is more reliable as a result.
What we're not claiming
To be direct about what COSMO isn't:
- Not autonomous AI — COSMO executes plans you define. It doesn't decide what to research on its own.
- Not a cognitive architecture — Earlier marketing used this language. It's really a task execution system with knowledge persistence.
- Not magic — It's software using LLMs to execute structured work. Results need verification like any AI output.
- Not finished — Active development. Things change.
The name
COSMO — Cognitive Orchestration System for Metacognitive Operations
"Cognitive orchestration" = coordinating multiple AI capabilities to do work.
"Metacognitive operations" = the system tracks what it's doing (task state, progress, dependencies).
The name made more sense when the architecture was more complex. It still fits, but "plan-driven task execution system" might be more accurate now.
Get in touch
If you're interested in COSMO—trying it, learning more, or discussing the approach—I'm happy to talk.
Email: jtr@evobrew.com
Website: cosmo.evobrew.com
Update: The COZ Journey
COSMO has been integrated with Claude Code and OpenClaw, taking the project on quite a journey through its development. What started as a task execution system evolved into something more—an exploration of AI coordination and persistent knowledge building.