FAQ and troubleshooting
Short answers to the things people ask most.
Is any of this real money?
No. The standalone payment adapter is a sandbox. Confirmed checkouts create order records and analytics, but no charge, refund, or live settlement occurs.
Do I need an API key?
No. Hosted models are included. If you'd rather use your own, run one locally - see Ollama & LM Studio.
What does a search cost?
Hosted runs store an estimated token/tool cost on the run record; the standalone build does not automatically charge it. Local runs use your own compute and are recorded with a zero platform cost.
Who can see my searches?
Only you. Runs are stored against your account and protected by row-level security. With a local model, the question never leaves your machine at all.
Can the assistant buy something I didn't approve?
No. Purchases require an explicit confirmation showing the price, currency and seller. In negotiations, your hard maximum is enforced in code, not by the model.
Why did my agent answer without searching?
Either its allowed tools list is empty, or the model you chose doesn't support tool calling. Check the agent's tool chips and, for local models, pick a tool-capable one.
The local model won't connect.
Nine times out of ten it's CORS. Add the app's exact origin (for example http://localhost:8080) to OLLAMA_ORIGINS and restart Ollama, or switch on Enable CORS in LM Studio's server settings.
My custom API tool returns nothing.
Custom tools are intentionally limited to public HTTPS GET endpoints. Run the agent and inspect its tool timeline for a guarded error or the bounded, untrusted response passed back to the model.
How do I create test profiles?
An authenticated admin can run the test-profile seeder from Admin → Seed. In production, the server must also have a TEST_ACCOUNT_PASSWORD configured.
Can teammates share my seller workspace?
Not in the standalone build. Seller data is owned by one user ID; team membership and shared-workspace roles are not implemented.