A plain list of what we shipped today, what we tested for real, what we have not tested yet, and our measured speed. Nothing here is a promise; the limits are in the open.
init, doctor (tells you what to fix next) and events (copy-paste signed examples in six languages), plus Docker and systemd files.We ran the agent against real services on 11 October 2026: a real Twilio account and phone number, a real Slack workspace and a real Google Calendar.
| Test | Result |
|---|---|
| Signed app event starts a workflow | Passed |
| Outbound phone call that plays the pack's recorded audio | Passed |
| Inbound call menu | Passed |
| Inbound text recognised and answered | Passed |
| STOP opt-out recorded; later texts and calls to that number blocked | Passed |
| Slack posts from three different workflows, and a signed slash command | Passed |
| Google Calendar (read-only): found an event, took the phone number and consent from its notes, fired the reminder workflow | Passed |
| Text delivered to a phone | Not proven: US carriers blocked it while our business-texting registration is still in review |
| Email, Microsoft Teams, Outlook, calendar event creation, GitHub and Stripe webhooks | Not tested yet |
The tests found four rough edges, and we fixed them the same day: the agent now records whether a text was really delivered (not just accepted), says plainly when a service is not set up, reports honestly when a text was blocked, and takes a single Slack channel setting. Behind those results are 85 automated tests. The voices are AI-generated, and non-English packs are marked native-speaker review pending.
These are our own medians from real calls, measured 9 October 2026 from a laptop in New York against our test environment, network time included. Full method and the numbers we are not proud of.
| What | Median |
|---|---|
| A sentence in an offline pack, or one we have already stored | 0.14 s (no generation, no per-play fee in a pack) |
| A brand-new short sentence, paid account (fast lane) | 0.37 s |
| A brand-new sentence, paid account | 0.48 s (0.80 s for a long, two-clause sentence) |
| First spoken sentence over our WebSocket, paid account | 0.54 s |
| A brand-new sentence, free account (standard lane) | 5.83 s |
The fast lane serves paying accounts: a new sentence comes back in about half a second, against about six seconds on the free lane. We are working on narrowing that gap for free accounts, and on getting faster still.