TechstarsSF

Talk to traces. Run simulations. Ship fixes.

The Problem

Your team has judgment about what the agent should do. Turning that judgment into evals and code fixes is hard. It's manual, inconsistent, and changes quickly.

The Solution

We turn failures into evals and fixes from one interface. Talk to your traces to capture your team's judgment and ship changes fast.

Get Started

Search traces and test prompt inputs

Fix incorrect outputs

John

Medium

Partial

Auto-tune prompts and tool calls

Prompt

Extract customer name

30 second setup

Into your repo with a few lines of code

terminal

$ poetry add simforge-py

main.py

from simforge import Simforge

client = Simforge(api_key=os.getenv("SIMFORGE_API_KEY"))
result = client.call("title-chat", user_query="What is the capital of France?")

Works where you work

Integrated into your IDE

VS Code

Cursor