# Current limits

This page lists what Connect cannot do today. It exists so that nobody discovers a limit during a customer conversation. Limits are grouped by channel, and each says whether it is a design decision, a provider dependency, or work that has not been done.

- **Status:** Available
- **Audience:** both
- **Last verified:** 2026-09-10
- **Canonical:** https://connectbyjbrh.com/docs/product/limits/

## Phone

| Limit | Kind | Detail |
|---|---|---|
| Call recording is not enabled | Provider capability | The capability is asked of the provider rather than assumed. It is not available on the live carrier, and no page here describes recording as though it were. |
| Human transfer is not completed end to end | Foundation | An escalation phrase queues a transfer and the supervisor panel can act on a live call, but a completed warm transfer depends on a provider capability that is not enabled. |
| Reply latency has a floor | Physics of the model | The measured floor on the realtime path is the model's first token plus its end-of-turn detection — a 3.3 s median on the best measured call, not the 1–2 s a demo video implies. |
| A call ends at thirty minutes | Design | The product's own ceiling, separate from the model session's limit. |
| Some speech behaviours cannot be steered | Model limit | Two were measured and rolled back rather than left in: respect-word cadence, and treating a murmur as a non-word. They are counted in the review and marked as model limits. |

## SMS

Inbound SMS, STOP handling and suppression work. Outbound SMS depends entirely on the configured provider, and the carrier on the live account carries no SMS at all. The Phone screen says so rather than presenting a thread that cannot send — which is the honest behaviour and also the reason SMS is listed as Foundation rather than Available.

In India, sending commercial SMS additionally requires DLT registration, which is a process with the telecom operators rather than a setting.

## Email

- A daily allowance bounds how much mail is processed; when it is spent, work is held rather than dropped, and the read cursor deliberately does not advance.
- Deliverability is not something any software can promise. Connect can hold the parts inside its control — authentication records, suppression, reply behaviour — and the rest belongs to your domain's history.
- A message is only ever reported as sent when the provider has acknowledged it. There is a third state, neither sent nor failed, and it is shown as uncertain rather than guessed either way.

## Prospecting

Connect does not guess email addresses. No pattern-generated address is ever produced, tested or sent to. A prospect with no discoverable address is kept as a researched prospect with a different next action, not converted into a guess. This costs volume and is not going to change.

Research is bounded by budget rather than by ambition: a deeper pass costs more and is only run where it can change a decision.

## The Assistant and the model

- The Assistant's rights are deliberately narrower than a person's. It cannot set a price, and it cannot clear a do-not-contact entry.
- An answer that Knowledge does not support is refused rather than improvised. That refusal is the correct behaviour and is not a bug report.
- Work that needs a paid model stops when the workspace's budget is exhausted. A caller already on the line still gets an answer; scheduled work waits.

## Sign-in and access

Google sign-in is the only route in. There is no password login, no email-and-link sign-in and no SSO with another identity provider.

## What is not on this page

Anything that is merely hard, slow or under construction is not called a limit here. If a capability is listed as Available elsewhere in this manual and it does not work, that is a defect — the [troubleshooting section](/docs/troubleshooting/) is the place to start, and reporting it is useful.

## Questions

### Why is call recording not available?

Because the provider capability is not enabled on the live carrier, and this system asks the provider what it can do rather than assuming. When it changes, this page and the capability status manifest change with it.

### Will Connect ever guess an email address?

No. It is a rule rather than a default, for compliance and reply-rate reasons that are set out in the engineering note on evidence-first prospecting.

### Is this list complete?

It is the complete list of limits that a business would notice. Per-capability detail sits on each capability page, and the machine-readable status of every capability is published at /docs-data/status.json.

## Related

- [Capability status](https://connectbyjbrh.com/docs/product/status/)
- [Connect by JBRH](https://connectbyjbrh.com/docs/product/)
- [Sending SMS: what is available today](https://connectbyjbrh.com/docs/sms/sms-outbound-status/)
- [Connect does not guess email addresses](https://connectbyjbrh.com/docs/prospects/no-guessed-email/)
- [What a setting cannot change](https://connectbyjbrh.com/docs/phone/model-limits/)

## What this page is based on

- AGENTS.md §9a — telephony capabilities and measured latency
- PROJECT-STATE.md — rolled-back voice experiments, 2026-09-06/07
- `backend/app/telephony.py` CAPABILITIES, `prospect_contacts.py`
- `docs-source/facts.py` FORBIDDEN_CLAIMS
