8000
Skip to content

examples

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

examples

README.md

GraphBit Examples

This directory contains ready-to-run Python scripts demonstrating various GraphBit workflow patterns using both local and cloud LLMs.

How to Run

  1. Set up your environment:
  • For local models (Ollama):
ollama serve
ollama pull llama3.2
  • For Perplexity (cloud):
export PERPLEXITY_API_KEY="your-api-key"
  1. Run an example:
python examples/tasks_examples/simple_task_local_model.py

🛠️ GraphBit Python API Tutorial

All example scripts in this directory use the GraphBit Python API to build and run AI workflows. Here’s a minimal step-by-step guide to the core GraphBit workflow pattern, as seen in these scripts:

1. Initialize GraphBit

from graphbit import LlmConfig, Executor, Workflow, Node

This sets up the GraphBit runtime and logging.

2. Configure Your LLM Provider

  • openai (cloud): 
    llm_config = LlmConfig.openai(model=gpt-3.5-turbo, api_key=api_key)
  • Ollama (local): 
    llm_config = LlmConfig.ollama("llama3.2")
  • Perplexity (cloud): 
    llm_config = LlmConfig.perplexity(api_key, "sonar")

3. Create an Executor

Choose the executor type based on your use case:

executor = Executor(llm_config, lightweight_mode=True)

# or for high-throughput pipelines:
executor = Executor(llm_config, timeout_seconds=60)

# or for memory-intensive tasks:
executor = Executor(llm_config, timeout_seconds=300)

# Configure additional settings for memory-intensive tasks if needed
executor.configure(timeout_seconds=300, max_retries=3, enable_metrics=True, debug=False)

4. Build a Workflow

Create a workflow and add agent nodes:

workflow = Workflow("My Example Workflow")
agent_id1 = str(uuid.uuid4())
node1 = Node.agent(
    name="Summarizer",
    prompt="Summarize: {input}",
    agent_id=agent_id1
)
agent_id2 = str(uuid.uuid4())
node2 = Node.agent(
    name="Task Executor",
    prompt="Summarize this text: {input}",
    agent_id=agent_id2
)
workflow.add_node(node1)
workflow.add_node(node2)
workflow.connect(node1,node2)
workflow.validate()

For multi-step or complex workflows, add multiple nodes and connect them as needed.

5. Run the Workflow

result = executor.execute(workflow)
if result.is_failed():
    print("Workflow failed:", result.state())
else:
    print("Output:", result.variables())

6. Example: Minimal End-to-End Script

from graphbit import LlmConfig, Executor, Workflow, Node
import uuid

llm_config = LlmConfig.ollama("llama3.2")
executor = Executor.new_low_latency(llm_config)
workflow = Workflow("Simple Task")
agent_id1 = str(uuid.uuid4())
node1 = Node.agent(
    name="Summarizer",
    prompt="Summarize: {input}",
    agent_id=agent_id1
)
agent_id2 = str(uuid.uuid4())
node2 = Node.agent(
    name="Task Executor",
    prompt="Summarize this text: {input}",
    agent_id=agent_id2
)
workflow.add_node(node1)
workflow.add_node(node2)
workflow.connect(node1,node2)
workflow.validate()
result = executor.execute(workflow)
print("Result:", result.variables())

Explore the scripts in this folder for more advanced patterns:

  • Real-time web search with Perplexity (simple_task_perplexity.py)
  • Memory-optimized large prompt tasks (memory_task_local_model.py)
  • Multi-step and dependency-based workflows (sequential_task_local_model.py, complex_workflow_local_model.py)
  • Conditional branching — personal financial advisor (balance intake + tiered advisors) (conditional_branch_local_model.py)

For more details, see the GraphBit Python API documentation.

Available Python Examples

simple_task_local_model.py
Single-agent workflow using the local Llama 3.2 model via Ollama. Summarizes a fictional journal entry.
Requires Ollama running locally.

sequential_task_local_model.py
Sequential multi-step pipeline using Llama 3.2. Each step addresses a different aspect of software IP protection, with outputs chained stepwise.
Requires Ollama running locally.

complex_workflow_local_model.py
Complex, multi-step workflow with explicit dependencies between tasks, covering a comprehensive IP protection strategy.
Requires Ollama running locally.

conditional_branch_local_model.py Condition node after an intake agent calls get_financial_details() (stub balance DEMO_ACCOUNT_BALANCE in the script); a Python handler routes by balance so exactly one of three advisors runs — Budget Planner (<$1k), Small Investment Advisor ($1k–$50k), Large Investment Advisor (>$50k) — each with different tools and tone. Loads .env for API keys; OpenAI if OPENAI_API_KEY else Ollama (tool calling is easier on OpenAI). Requires either OPENAI_API_KEY or Ollama.

memory_task_local_model.py
Memory-intensive, single-agent task with a large prompt, using Llama 3.2 via Ollama. Provides a deep legal/technical analysis.
Requires Ollama running locally.

simple_task_perplexity.py
Single-agent workflow using Perplexity’s cloud models (with real-time web search). Summarizes recent AI/ML developments.
Requires PERPLEXITY_API_KEY environment variable.

chatbot
A conversational AI chatbot with vector database integration for context retrieval and memory storage. Includes a FastAPI backend and Streamlit frontend.
Requires OpenAI API key and ChromaDB.

llm_guided_browser_automation.py
Automates browser interactions using LLMs to guide actions. Demonstrates how to use GraphBit for real-time decision-making in web automation tasks.
Requires Selenium and a configured LLM provider.
0