AGNT5 Documentation

Durable AI agents that survive failures

Developer quickstart

Build your first durable workflow in minutes.
Learn the basics of the AGNT5 platform.

Get Started

from agnt5 import workflow, WorkflowContext

@workflow
async def my_workflow(ctx: WorkflowContext):
    result = await ctx.step("fetch", fetch_data)
    return result

Build a chatbot Create an agent with memory and conversation history

Deploy to production Ship your workflows to staging and production environments

Integrations Connect to LLM providers, vector stores, and external services

Workflows Multi-step orchestration with built-in checkpointing and recovery

Build

Define agents, workflows, and functions in code with built-in durability

Deploy

Go from local dev server to production with one command

Iterate

Debug with traces, replay failures, and observe behavior in Dev Studio

How to build agents and workflows

Define agents, workflows, and functions in code. Everything is durable by default.

Goal	What to use	Description
Build an AI agent with capabilities	Agent, @tool	LLM-powered agent with tool calling. Tools let agents interact with external systems.
Create durable workflows	@workflow, @function	Multi-step orchestration with checkpointing. Functions are the atomic building blocks.
Add memory to agents	ConversationMemory, SemanticMemory	Persistent chat history and vector search over documents.

How to deploy to production

Run locally with SQLite, deploy to production with PostgreSQL. One command each.

Goal	What to use	Description
Go from local to production	agnt5 deploy	Deploy to cloud with one command.
Self-host or scale	Community Edition, Managed Edition	PostgreSQL for teams, or event-sourced architecture for high-throughput.
Manage secrets	agnt5 secrets	Store API keys securely. Injected at runtime, never in code.

How to debug and iterate

Dev Studio provides real-time visibility into execution. No instrumentation required.

Goal	What to use	Description
Observe and debug	Execution timeline, Trace viewer	Visualize agent steps in real-time. Inspect traces with latency breakdown.
Test and iterate	Dev Studio UI, Replay	Run workflows from browser. Re-run historical executions with updated code.
Add oversight and monitoring	Human-in-the-loop, Logs & metrics	Pause for review, filter logs by severity and trace ID.