PRAISONAI_DOCS = """
# PraisonAI Agent Framework — Full Documentation

## What is PraisonAI?
PraisonAI is a multi-agent orchestration framework. You (the orchestrator) can create specialized
sub-agents, each with custom tools written in Python, to complete complex tasks collaboratively.

---

## Creating an Agent
```python
from praisonaiagents import Agent, Task, PraisonAIAgents

agent = Agent(
    name="ResearchAgent",
    role="Research Specialist",
    goal="Find and summarize info",
    backstory="Expert researcher",
    llm="LongCat-Flash-Lite",
    tools=[my_tool_fn],
    verbose=True,
)
```

---

## Creating a Task
```python
task = Task(
    name="research_task",
    description="Research the latest developments in quantum computing",
    expected_output="A detailed summary with key findings",
    agent=agent,
    context=[previous_task],
)
```

---

## Creating Custom Tools
Tools are plain Python functions. Use inline comments (# not triple quotes) inside tools.

Example tools:
```python
def search_web(query: str) -> str:
    # Search the internet for information.
    import requests
    return results

def execute_python(code: str) -> str:
    # Execute Python code and return output.
    import subprocess
    result = subprocess.run(['python3', '-c', code], capture_output=True, text=True)
    return result.stdout + result.stderr

def calculate(expression: str) -> str:
    # Evaluate a mathematical expression.
    return str(eval(expression))

def read_file(filepath: str) -> str:
    # Read content from a file.
    with open(filepath) as f:
        return f.read()
```

---

## Running Multiple Agents (Sequential)
```python
system = PraisonAIAgents(
    agents=[researcher, analyst, writer],
    tasks=[research_task, analysis_task, write_task],
    process="sequential",
    verbose=True
)
result = system.start()
```

---

## Running Multiple Agents (Hierarchical)
```python
system = PraisonAIAgents(
    agents=[manager_agent, worker1, worker2],
    tasks=[main_task, sub_task1, sub_task2],
    process="hierarchical",
    manager_llm="LongCat-Flash-Lite",
    verbose=True
)
result = system.start()
```

---

## Task Context (Chaining)
```python
task1 = Task(name="t1", description="...", agent=agent1, expected_output="...")
task2 = Task(name="t2", description="...", agent=agent2, expected_output="...", context=[task1])
```

---

## Best Practices for Dynamic Tool Creation
1. Use snake_case names (e.g., fetch_github_repo, parse_csv_data)
2. Always include type hints on parameters and return value
3. Use inline # comments, NOT triple-quoted docstrings inside tool code
4. Wrap risky operations in try/except
5. Always return strings

---

## Sub-Agent Design Patterns

Pattern 1 - Research + Synthesis:
  Agent 1 (Researcher): Gathers raw information
  Agent 2 (Analyst): Processes and analyzes data
  Agent 3 (Writer): Produces final output

Pattern 2 - Divide and Conquer:
  Split large tasks, each agent handles one domain, orchestrator merges results

Pattern 3 - Validation Pipeline:
  Agent 1: Main task, Agent 2: Validates, Agent 3: Applies corrections

---

## LongCat-Flash-Lite Configuration
- Model ID: LongCat-Flash-Lite
- Max tokens: 327,680 (320K context)
- Speed: Very fast (~500-700 tokens/sec)
- Free quota: 50M tokens/day
- API Base: https://api.longcat.chat/openai/v1
- Compatible with OpenAI SDK format
"""