Many sides.
In your chat.
Explore an idea with a simulated audience in Codex or Claude Code.
1. Connect your agent
You’ll need the project checkout, Python 3.10 or newer, and your client’s command-line tool. Run one command from the project folder.
Need the project? Create an account or contact us for access.
python3 scripts/setup_chat.py --client codexpython3 scripts/setup_chat.py --client claudeSetup installs the local MCP server and companion skill into your client. Keep the project folder in place, then start a new chat.
2. Try a preview
Use the installed audience-review tools to run an offline demo with 20 simulated reviewers in Alameda County. My idea: a grocery app that shares lists between housemates.The default is an offline demonstration with arbitrary sample answers. It shows how the workflow works; it is not audience feedback. Live synthetic reviews require separate local configuration and explicit authorization for paid calls. Website reviews require sign-in and draw from your account’s free AI credit.
3. Follow the evidence
- “Show the individual responses behind the appeal score.”
- “Show my saved reviews.”
- “Compare this revision using the same audience, seed and questions: …”
Your agent can select a supported county, follow a review running in the background, retrieve saved history and inspect individual response records. Availability depends on the connected local server.
Setup help
Add --dry-run to a setup command to preview its changes. Setup preserves an existing server registration or modified skill instead of overwriting it. If your client is not found, install its CLI first. For live setup and troubleshooting, use docs/chat-setup.md in your checkout. Keep credentials out of chat.
Instructions for your agent
Give your agent the plain-text guide. It covers tool discovery, supported audiences, review progress and evidence lookup.