Alphaproduct
voice-of-agents
Beta users lie. Simulate them honestly.
An MIT-licensed Python library that simulates user research with Claude. It hands the resulting personas to a live browser-based eval harness that navigates your real product. Published on PyPI.
- MIT licensed
- Published on PyPI
- Offline cassette mode
- Alpha
pip install voice-of-agents- Agentic orchestration
- Data science & evaluation
- NLP & information extraction
- test files, approximately
- 34
- pipeline stages
- 4
The problem
User research before product-market fit is too slow and expensive. Founders ship by instinct and learn why users churn only after they have churned.
The approach
The library runs four stages: hypotheses and subjects, typed personas, workflows, and journey redesign. The personas then feed a live browser eval harness that drives your actual product, which makes it more than a prompt library. It is not a replacement for real research. It is a forcing function for better questions.
How voice-of-agents fits together
Hypotheses
Research framing and subjects
Personas
Typed, with pains and capabilities
Workflows
And the gaps between them
Journey redesign
Live eval
A browser driver runs the personas
What voice-of-agents does
Simulate them honestly
Beta users lie. Simulate them honestly, then validate the risky findings with the users you still need to talk to.
Feeds a live browser eval
Beyond answering prompts, the personas drive a real eval harness against your actual product.
Adopters, abandoners, skeptics, critics
The sampling frame mandates all four, so the happy path can't dominate by construction.
A forcing function, not a replacement
It is not a replacement for real research. It is a forcing function for better questions.
What’s inside voice-of-agents
- research/ module: hypotheses and subject framing
- design/ module: persona and workflow pipelines, gap analysis
- eval/ module: browser driver, the research-to-eval bridge, run diffing
- core/ module: persona, pain, capability and backlog models
- cli/ module: the voa command-line entry point
Beta users lie. Simulate them honestly, then validate the risky findings with the users you still need to talk to.
Where voice-of-agents stands today
- Published on PyPI with CI and trusted publishing
- Roughly 34 test files, a recorded end-to-end run, and a non-suppressible synthetic-data notice emitted every session
- An offline cassette mode, so the demo runs without credentials
The voice-of-agents tech stack
- Python 3.11+
- Click
- Playwright
- Pydantic
- Anthropic SDK
- hatchling
Install voice-of-agents
pip install voice-of-agents