🚀 Cogency is live! Built emergent AI reasoning in 9 hours - looking for feedback

[iteebz]

Hey everyone! 👋

I just shipped Cogency after a 9-hour build session, and I'm excited to share it with the community. This started as a personal project to escape LangChain boilerplate hell, but it turned into something much more interesting.

What I Built

A cognitive architecture that demonstrates emergent reasoning behavior through clean abstractions. The core insight: separating cognitive operations from tool orchestration enables true composability.

Architecture: Plan → Reason → Act → Reflect → Respond

The Emergent Behavior

When I added WebSearchTool to test extensibility, something unexpected happened:

• Agent made initial search with poor query
• Got irrelevant results
• Reflected on the failure
• Adapted query strategy automatically
• Succeeded on retry

This wasn't programmed behavior—it emerged from the architecture's separation of concerns.

Current State

• ✅ Production-ready (pip installable)
• ✅ 115+ unit tests
• ✅ Built-in tools: Calculator, WebSearch, FileManager
• ✅ Clean tracing and debugging
• ✅ Auto-discovery of custom tools

What I'm Looking For

Feedback on:

• The cognitive architecture approach
• Tool composition patterns you've seen work
• Edge cases I should test
• Features that would be most valuable

Contributions welcome:

• New tool implementations
• LLM integrations beyond Gemini
• Performance optimizations
• Documentation improvements

Try It Out

pip install cogency

from cogency.agent import Agent
from cogency.llm import GeminiLLM
from cogency.tools import CalculatorTool, WebSearchTool

agent = Agent(
    name="TestAgent",
    llm=GeminiLLM(api_key="your-key"),
    tools=[CalculatorTool(), WebSearchTool()]
)

result = agent.run("What is 15 * 23?", enable_trace=True)

Discussion Questions

1. Architecture: What patterns have you seen for emergent vs programmed behavior in agent systems?
2. Tools: What tools would be most valuable to add next? I'm considering:

• Database connectors
• API clients
• Code execution environments
• Image processing tools

3. LLM Support: Priority order for additional LLM integrations?
4. Performance: Any thoughts on optimizing the reasoning loop?
5. Use Cases: What problems are you trying to solve with agent frameworks?

Roadmap

Short-term:

• More LLM integrations (Claude, GPT-4)
• Additional built-in tools
• Performance benchmarks
• JavaScript implementation

Long-term:

• Multi-agent coordination
• Memory and state management
• Visual reasoning loop editor
• Plugin ecosystem

Get Involved

• 🐛 Found a bug? Open an issue
• 💡 Have an idea? Start a discussion
• 🛠️ Want to contribute? Check out the contributing guidelines
• 📖 Documentation improvements always welcome

The commit timeline tells the full story of the 9-hour build. Looking forward to hearing your thoughts and seeing what you build with it!