AI Agents
Build intelligent agents that reason, use tools, and accomplish complex tasks with built-in durability
AI Agents
Learn how to build, deploy, and monitor AI agents with AGNT5.
Agents are AI systems that autonomously accomplish tasks—from answering questions with tools to executing complex multi-step workflows. AGNT5 provides the SDK to build agents, infrastructure to run them reliably, and Dev Studio to monitor and debug them.
The Platform
AGNT5 is a platform for building production AI agents with built-in durability, observability, and scale.
Build
Define agents in code with tools, memory, and multi-agent orchestration
Deploy
Scale from local dev server to distributed production with one command
Monitor
Debug with traces, replay failures, and observe agent behavior in Dev Studio
How to build an agent
Building an agent starts with defining its capabilities in code. Connect LLMs, equip them with tools, and let the agent handle conversation history automatically.
| Goal | What to use | Description |
|---|---|---|
| Create an AI agent | Agent | LLM-powered agent with tool calling and automatic conversation history |
| Give agents capabilities | @tool | Functions the agent can call, with automatic schema generation |
| Coordinate multiple agents | handoffs | Transfer control between specialized agents |
Quick example
from agnt5 import Agent, tool
@tool
async def search_docs(ctx, query: str) -> list[str]:
"""Search the documentation for relevant information."""
return await vector_store.search(query, limit=5)
@tool
async def create_ticket(ctx, title: str, description: str) -> str:
"""Create a support ticket in the system."""
return await ticketing_api.create(title=title, description=description)
support_agent = Agent(
name="support",
model="anthropic/claude-sonnet-4-20250514",
instructions="You help users with technical questions. Search docs first, create tickets for bugs.",
tools=[search_docs, create_ticket],
)
# Run the agent (streaming)
async for event in support_agent.run("How do I configure retry policies?"):
if event.event_type == "lm.content_block.delta":
print(event.content, end="")
# Or run without streaming
result = await support_agent.run_sync("How do I configure retry policies?")
print(result.output)Deploy to production
AGNT5 runs anywhere—from a local SQLite dev server to distributed Kubernetes clusters.
| Mode | Backend | Use case |
|---|---|---|
| Embedded | SQLite | Local development, prototyping |
| Community | PostgreSQL | Self-hosted production, team deployments |
| Managed | PostgreSQL + event log | High-throughput, horizontal scaling |
# Deploy to production
agnt5 auth login
agnt5 deployMonitor and debug
Dev Studio provides real-time visibility into agent execution.
| Feature | Description |
|---|---|
| Execution timeline | Visualize agent steps, tool calls, and LLM interactions |
| Trace viewer | Inspect distributed traces with latency breakdown |
| Logs & metrics | Filter by severity, trace ID, and agent name |
| Interactive testing | Run agents and workflows directly from the UI |
| Human-in-the-loop | Approve, reject, or provide input to paused executions |
| Replay | Re-run historical executions with updated code or prompts |