Skip to main content

​
What is Simforge?

Simforge helps you understand what your AI is doing and discover what’s going wrong. You add tracing with our SDK to capture enough context to see when and why calls fail, then our agent builds automated graders to continuously check for those issues.
1

Capture

Add the SDK to capture inputs, outputs, errors, and context for every AI call — enough to tell whether a call succeeded or failed, and why. 1-shot integrations with the MCP.
2

Describe

Ask our agent to find failures for you or describe issues you’ve encountered in plain english.
3

Improve

Our agent suggests fixes, trains graders to evaluate future failures, and suggests code fixes to your coding agents.

​
Quick Start

Get API Keys

Create API keys to authenticate with Simforge

TypeScript SDK

Add tracing in TypeScript

Python SDK

Add tracing in Python

Ruby SDK

Add tracing in Ruby

Go SDK

Add tracing in Go

MCP Setup

Let your coding agent set up tracing automatically

​
Add Tracing

Wrap your AI functions to start capturing traces. Here’s all it takes: 
import { Simforge } from "@goharvest/simforge"

const simforge = new Simforge({ apiKey: process.env.SIMFORGE_API_KEY })
const myService = simforge.getFunction("my-service")

const tracedFn = myService.withSpan(async (text: string) => {
  return { result: "..." }
})
Once traces are flowing, open Simforge and start a conversation with the grading agent to ask for open ended analysis or describe problems you see with your agents or LLM workflows. It handles the rest.