Get started AGNT5 Foundations

AGNT5 Foundations

Four building blocks for reliable AI systems

Four building blocks for reliable AI workflows. Each solves a different problem.

The Four Primitives

Function Entity Workflow Agent
What Stateless operation Stateful component Multi-step process AI-powered system
State None Persistent per key Per workflow run Conversation history
Recovery Automatic retries State survives restarts Resume from checkpoint Context preserved
Best For API calls, embeddings User sessions, memory Pipelines, orchestration Reasoning, decisions
Example Generate embedding Chat history RAG pipeline Support agent

When to Use What

Need AI to make decisions? → Use Agent with tools

Need multiple coordinated steps? → Use Workflow to orchestrate them

Need to remember state across calls? → Use Entity with a unique key

Just execute and return? → Use Function

Common Patterns

Pattern 1: AI Chatbot 

Entity (user session) + Agent (chat)

Entity maintains conversation history, Agent handles responses.

Pattern 2: Content Generation Pipeline 

Workflow orchestrates:
  ├─ Function: Retrieve documents
  ├─ Agent: Generate content
  └─ Function: Publish result

Workflow ensures pipeline completes even if steps fail.

Pattern 3: Autonomous Assistant 

Workflow + Agent + Tools

Workflow manages overall flow, Agent reasons and uses Tools to accomplish goals.

Quick Example 

from agnt5 import Agent, workflow, tool, Context, WorkflowContext

# Tool for the agent
@tool
async def search_docs(ctx: Context, query: str) -> str:
    """Search documentation for answers."""
    return f"Found results for: {query}"

# Agent with tools
agent = Agent(
    name="assistant",
    model="openai/gpt-4o-mini",
    instructions="You are a helpful assistant.",
    tools=[search_docs]
)

# Workflow orchestrating agent
@workflow()
async def process_question(ctx: WorkflowContext, question: str) -> dict:
    """Durable workflow with checkpointing."""

    # Step 1: Get answer from agent (checkpointed)
    result = await ctx.step("get_answer", agent.run(question))

    # Step 2: Store result in workflow state
    ctx.state.set("last_question", question)
    ctx.state.set("last_answer", result.output)

    return {"question": question, "answer": result.output}

# If crash happens after step 1, resumes from step 2

Next Steps

Core Primitives

  • Functions - Stateless operations with retries
  • Entities - Stateful components with unique keys
  • Workflows - Multi-step orchestration patterns
  • Agents - Autonomous LLM-driven systems