Advanced AI Chain of Thought Analysis

What This System Does

The Chain of Thought Analysis System mimics human-like reasoning by:

• Analyzing query complexity before planning
• Generating multiple candidate plans for solving problems
• Evaluating and selecting the best plan
• Executing tasks sequentially with immediate quality evaluation
• Iterating when results don't meet quality thresholds
• Synthesizing findings into coherent final answers

Design Philosophy

Helping Smaller Models Punch Above Their Weight

This system is designed to help smaller language models deliver analysis far more sophisticated than what one might expect from their size alone. By breaking down complex problems into manageable pieces, structuring the reasoning process, and enforcing quality checks at each step, the system enables models like GPT-OSS 20B to perform analysis that would normally require much larger models.

Controlled Reasoning

Models like GPT-OSS 20B and 120B include their own internal reasoning capabilities. This program disables that internal reasoning as much as possible because it provides its own structured reasoning framework. The goal is to shift the reasoning process onto the program's architecture, where it can be controlled, monitored, and quality-checked at each step.

Three Core Components

Three Core Components Architecture

How It Works

%%{init: {'theme': 'base', 'themeVariables': { 'primaryColor': '#1a1a2e', 'primaryTextColor': '#fff', 'primaryBorderColor': '#4a4a6a', 'lineColor': '#6c63ff', 'secondaryColor': '#16213e', 'tertiaryColor': '#0f3460', 'fontFamily': 'JetBrains Mono, monospace'}}}%%
flowchart TD
    subgraph INPUT["📥 INPUT"]
        A[/"User Query"/]
    end

    subgraph ANALYSIS["🔍 ANALYSIS"]
        B["1. Complexity Analysis"]
        C["2. Plan Generation"]
        D["3. Plan Evaluation & Selection"]
    end

    subgraph EXECUTION["⚡ EXECUTION"]
        E["4. Task Execution"]
        F{"5. Iteration
Decision"}
    end

    subgraph OUTPUT["📤 OUTPUT"]
        G["6. Synthesis"]
        H["7. Focused Answer"]
        I[/"Final Result"/]
    end

    A --> B
    B --> C
    C --> D
    D --> E
    E --> F
    F -->|"Continue"| B
    F -->|"Done"| G
    G --> H
    H --> I

    style A fill:#4a4a6a,stroke:#6c63ff,stroke-width:2px,color:#fff
    style I fill:#4a4a6a,stroke:#6c63ff,stroke-width:2px,color:#fff
    style F fill:#0f3460,stroke:#e94560,stroke-width:2px,color:#fff
    style B fill:#1a1a2e,stroke:#4a4a6a,color:#fff
    style C fill:#1a1a2e,stroke:#4a4a6a,color:#fff
    style D fill:#1a1a2e,stroke:#4a4a6a,color:#fff
    style E fill:#1a1a2e,stroke:#4a4a6a,color:#fff
    style G fill:#1a1a2e,stroke:#4a4a6a,color:#fff
    style H fill:#1a1a2e,stroke:#4a4a6a,color:#fff

Each task is evaluated immediately after completion. Only accepted tasks write their results to analysis history, ensuring subsequent tasks see only quality results.

Input Limits

These are character limits, not token limits.

Best Practice for Context

Rather than feeding in massive documents and hoping the LLM finds what's relevant, pre-process and focus your context:

1. Extract relevant sections that directly relate to your query
2. Summarize background information to its essential points
3. Remove redundant or tangential content
4. Use another tool (or this program itself) to distill large documents first

Quick Start

Simple One-Shot Analysis

import asyncio
from agent import run_analysis

async def main():
    result = await run_analysis(
        query="What are the key factors affecting renewable energy adoption?",
        context="Focus on economic and policy factors in the United States.",
        groq_api_key="your-groq-api-key",
        params={
            "mode": "balanced",
            "max_iterations": 3,
        }
    )
    
    print(f"Final Answer: {result.final_answer}")
    print(f"Execution Time: {result.execution_time:.2f}s")
    print(f"Success: {result.success}")

asyncio.run(main())

With More Control

import asyncio
from agent import create_agent

async def main():
    agent = await create_agent(
        agent_id="myagnt",
        groq_api_key="your-groq-api-key",
        params={
            "model": "openai/gpt-oss-120b",
            "mode": "thorough",
            "cost_limit": 0.50,
        }
    )
    
    try:
        result = await agent.analyze(
            query="Analyze the potential impact of quantum computing on cryptography.",
            context="Consider both near-term and long-term implications."
        )
        
        print(f"Success: {result.success}")
        print(f"Final Answer: {result.final_answer}")
        
    finally:
        await agent.cleanup()

asyncio.run(main())

Example Output

After running an analysis, you receive a MainAnalysisOutput object:

result = await run_analysis(
    query="What factors affect solar panel efficiency?",
    groq_api_key=key,
    params={"focused_answer_type": "number"}
)

# Access different answer layers
full_synthesis = result.synthesis_display.answer      # Comprehensive analysis
final_answer = result.final_answer_display.answer     # Refined, user-ready response
focused_value = result.focused_answer_display.answer_value  # Constrained answer

# Metadata
print(f"Title: {result.title}")
print(f"Execution time: {result.execution_time:.2f}s")
print(f"Tasks executed: {result.task_count}")
print(f"Iterations: {result.total_iterations}")

# Cost tracking
print(f"Total cost: ${result.cost_snapshot['costs']['total_cost']:.4f}")

Available Models

Both models have 128K token context windows and are available under Apache 2.0 license.

Execution Modes

Requirements

• Groq API Key (required) — For LLM access
• Tavily API Key (optional) — Enables real-time news search
• Replicate API Key (optional) — Enables AI image generation for reports

Documentation Pages