Post-training platform
Send your agent’s conversations. Simulate the ones that fail. Score every one. Train on what moved the number, with a test set kept apart.
$ pip install zeroproof
import zeroproof.simulations as zps# 1. simulate: tools + prompt become a worlddata = zps.simulate(tools=my_tools, system_prompt=my_prompt, mode="rl")# 2. score: your rules, every conversationscored = data.grade(judge=my_judge)scored.pass_at # how often it passes# 3. train: keep what teaches, publishrows, report = zps.optimize(data, mode="rl")data.push("support-v1", agent="airline-support", holdout=0.1)
Platform
Traces
Three settings in your app. Every conversation it has in production lands here, scored on the behaviors you name.
OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=https://api.zeroproofai.com/v1/traces
Simulate
Your agent’s tools and prompt become a world. Customers, broken tools, thousands of conversations, aimed at what fails in production.
zps.simulate(tools=my_tools, system_prompt=my_prompt)
Score
Your rules, applied to every conversation. Three numbers: how often it passes, how often it passes every time, how much training can still gain.
scored = data.grade(judge=my_judge)
Train
Only the conversations a model can learn from, with a test set kept apart. One line to pull into any trainer.
rows = zps.pull("ds_…")
Results
Open 8B model, customer-support benchmark
tasks passed
Same number of examples. None written by a person. Method
A 1B model that checks refund requests
right answers vs cost
Trained inside a simulated marketplace. Model, data and test set public. Method
Datasets
Support agents, coding agents, and yours. Rows and pass rate on every card. A test set with each one.
Research
Simulated data pipeline for fine-tuning
A $15 simulated training set beat expert-written examples on a public support benchmark.
Safety judges for agentic commerce
A 1B refund checker trained inside a simulated marketplace.
How training data gets made
One training example, step by step, on a real airline agent.
pip install zeroproof && zeroproof signup --email you@company.com