Skip to content

FastAIAgent SDK

Build, debug, evaluate, and operate AI agents.

The only SDK with Agent Replay — fork-and-rerun debugging for AI agents.

Works standalone or connected to the FastAIAgent Platform for visual editing, production monitoring, and team collaboration.

What Makes FastAIAgent Different

Feature FastAIAgent LangSmith Langfuse
Agent Replay (fork-and-rerun) Yes No No
Durable HITL — pause for days, resume from any process Yes No No
@idempotent side-effect protection Yes No No
Build agents in code Yes No No
Cyclic chain workflows Yes LangGraph No
Multi-turn agent simulation Yes No No
Built-in guardrails Yes No No
OpenAI Decisions API — route, guard and judge with probabilities, replayable Yes No No
Safety library (PII, prompt-injection, moderation) Yes No No
OTel-native tracing Yes Proprietary Proprietary
Fragment prompt composition Yes No No
Visual editor sync Yes No No

Quick Start

pip install fastaiagent

from fastaiagent import Agent, FunctionTool, LLMClient

agent = Agent(
    name="assistant",
    system_prompt="You are a helpful assistant.",
    llm=LLMClient(provider="openai", model="gpt-4o-mini"),
    tools=[FunctionTool(name="greet", fn=lambda name: f"Hello, {name}!")]
)

result = agent.run("Say hello to World", trace=True)
print(result.output)
print(result.trace.summary())

Core Features

  • Agents — Build agents with tools, memory, and multi-agent teams
  • Durability — Pause for human approval, survive crashes, resume from any process. SQLite locally, Postgres in production.
  • Streaming — Real-time token delivery from LLM to your app
  • Structured Output — Force LLM responses into typed JSON schemas
  • OpenAI Decisions API — Fixed-answer questions that return probabilities, not text. They power guardrails, an eval judge, Chain routing, an agent tool, and a Supervisor that routes with the Decisions API (call-centre walkthrough)
  • Chains — Directed graph workflows with cycles, typed state, and checkpointing
  • Guardrails — Input/output/tool validation (code, regex, LLM judge)
  • Tracing — OTel-native tracing with local SQLite storage
  • Agent Replay — Fork-and-rerun debugging at any execution step
  • Evaluation — Scorers, datasets, LLM-as-judge, trajectory eval
  • Prompts — Registry with versioning and fragment composition
  • Knowledge Base — Local file ingestion with embedding search, or PlatformKB for hosted KBs
  • Platform Connection — Connect for traces, prompts, eval, and replay
  • Integrations — Auto-tracing for OpenAI, Anthropic, LangChain, CrewAI

Next Steps