What's new in Orbit

The WhatsApp test-account verification card used to validate the number with a formatting-tolerant check, then send the literal typed string. That let "+1 (234) 567-8900" pass the local check but gets rejected by the server's strict E.164 schema with a 400, so the "Send Code" button felt broken and users rageclicked it. The UI now canonicalizes the phone (strips spaces, dashes, parens, dots) before calling send-otp, so the endpoint accepts what the UI accepted.

**What changed** — Two more dashboard surfaces opened their Server-Sent-Events streams on a relative path. The browser resolves those against the dashboard origin, which has no API route, so each stream disconnected on 404s and the panel simply never populated — no error, just silence. They now resolve through the same API-base helper every other SSE consumer uses:

- **Supervisor coach panel (live wallboard)** — the oversight card with the latest supervisor-assist recommendation and live compliance findings for the selected call.
- **Agent Studio test-by-voice** — the live partial transcript while you speak-test an agent.

After a page reload, both panels show content as it streams in, instead of a permanently empty surface.

**What's new** — The [voice AI agents](/voice-ai-agents) product page now answers the question buyers actually ask when shortlisting: how does a voice AI agent integrate into the business workflows and systems we already run? The answer walks through the integration points — grounding answers in your existing help content through the knowledge base, calling your live systems mid-call through registered tools (order lookup, slot booking, ticket creation), writing post-call summaries and action items back to your CRM or webhook endpoints, and escalating to your team in the same inbox that handles SMS, WhatsApp, RCS, and email.

**Why it matters** — Evaluation pages that only describe capabilities leave the integration question unanswered until a sales call. The [voice AI agents](/voice-ai-agents) and [WhatsApp Business API](/whatsapp-business-api) pages also now link to each other, so a buyer evaluating the combined voice-and-messaging agent setup can move between the two product pages directly.

**What changed** — The [Best Twilio Alternatives for Voice, SMS, and AI Agents 2026](https://orbit.devotel.io/en/compare/best/twilio-alternatives-voice-sms-ai-agents-2026) round-up closes its short-answer summary with a bottom-line verdict: for the voice + SMS + AI agents requirement, an omnichannel platform qualifies, a focused AI-voice tool does not — and published pay-as-you-go rates per destination let buyers compare the assembled multi-vendor cost side by side against one bill. The earlier closing sentence described the page itself; the rankings, vendor list, and feature table are unchanged.

One provider metadata field drifted from the platform configuration spec, and
the two could disagree when the messaging provider was probed. This fix brings
the field back in sync with the authoritative configuration layer so every
provider check reports the same value.

The [Connect via SMPP](/guides/smpp) guide covers how to bind your client and send. A new [SMPP edge model](/concepts/smpp-edge-model) concepts page now documents what sits underneath the bind: each credential you create maps to one user on the SMPP relay, a reconciliation service applies your credential and carrier configuration every ~30 seconds (so a status change such as suspend or revoke takes effect within a tick rather than instantly), your credential's DLR mode decides whether delivery receipts return on the bind or POST to a webhook, and bring-your-own carrier routing follows a declarative scope/priority model. A troubleshooting table maps bind vs. submit_sm failures to whether the credential, your network, or the routing configuration owns the fix.

SMPP traffic depends on the smpp_bind_owners lookup table, and the
platform's post-startup self-heal task is what guarantees it exists. A
transient database connectivity blip during that task's catalog check
previously wedged the table's guaranteed-recovery permanently. The task now
treats an unreadable catalog as "assume absent" and still runs the
table-creation step safely, so SMPP-dependent flows recover instead of
failing on every boot after a blip.

**What changed** — The comparison hub at /compare gains a new page,
**Alternatives for Small Business Communications 2026**, that weighs Orbit
against Twilio, RingCentral, Vonage, and Dialpad specifically for a small
business. It flips the usual checklist to the two questions that decide it:
who owns the carrier network your calls and messages ride on, and whether the
AI voice agent is built in or bolted on. Orbit ships both — its own wholesale
switch (carrier of record) and native AI agents — so the page evaluates each
provider against that bar. The page is also linked from the hub's
Small Business segment below, so shoppers and AI answer engines can find it.

When you move an open quality evaluation from one evaluator to another, the Reassign dialog lists every eligible evaluator with their current load. If the request that builds that list failed, the dialog used to render an empty picker offering only the "Select an evaluator" placeholder — nothing conveyed what went wrong, and the only way out was to close the dialog or reload the page.

The dialog now detects that failure and shows a Retry button alongside an explanation. Clicking Retry re-fetches the evaluator list; once it succeeds you pick a new evaluator and reassign as normal. The Reassign button also stays disabled while the list is in this failed state, so a submission cannot fire against an incomplete picker.

**What changed** — The dedicated push send endpoint used by the dashboard's Send Push dialog previously wrote its per-device delivery outcomes only to the legacy push log. The unified Messages view for the push channel reads the shared message log, so anything sent through the dedicated route was invisible there until a send went through the standard messaging API instead. The endpoint now mirrors every outcome (sent, failed, and frequency-capped skips) into the shared log as well, once per targeted device, while keeping the legacy log updated as before. Delivery behavior and the legacy push management pages are unchanged; a mirrored log entry that cannot be written never delays or fails the send itself.

**A dedicated guide for propensity-driven segments** — Devotel Orbit's Audience pages let you rank contacts by a trained churn-propensity or lifetime-value model and save the top scorers as a segment, but until now the docs covered only rule-based segments and the predictive models themselves. The new [propensity segments guide](https://docs.orbit.devotel.io/guides/propensity-segments) closes that gap: it explains how a propensity segment differs from a filter-driven segment, what has to be true before you can save one, and how the segment stays current.

The guide walks the full loop — train the model, score the population to preview the ranking, activate the top-N into a saved segment, target it in a campaign or export it, and re-materialize it on the per-model activation schedule rather than editing the segment by hand. It also covers the failure modes a customer actually hits: an under-trained model, an empty scoring population, and a segment frozen against a stale fit.

**What's new** — The [pre-send policy scanner guide](/docs/compliance/policy-scanner) now documents the control that checks every outbound message before it dispatches: TCPA quiet hours, SHAFT content, GDPR sender identity, spam-keyword scoring, and DLP (credit cards, Social Security numbers, IBANs, opt-in passport). It explains the organization-wide scan mode — strict blocks a flagged send, warn records it, off skips scanning — with curl examples for the mode endpoint, and shows how POLICY_VIOLATION and the retryable lookup-failure (503) surface in send responses.

**Why it matters** — Compliance posture used to live only in the API reference row and error-code list. Teams setting up their preferred enforcement level, or building the compose-time live lint into a workflow, now have a dedicated guide that covers the DLP privacy model (offsets, never text), the checksum validation that keeps false positives low, and the tenant-owned nature of the control.

**What changed** — Users who have picked a non-English locale on the signup
form or settings page used to receive the password-reset email in English
even though the welcome email was translated. The password-reset template
now ships translated copy across the product's six locales (English,
Turkish, Arabic, German, French, Spanish), with English as the fallback
for any unrecognized tag, matching the welcome email behaviour.

**What's new** — [Platform as a service for communications: what must be true before you build on it](/paas-infrastructure) is an evaluation guide for businesses weighing what their communication stack is built on. It defines the PaaS model in plain terms, lays out a six-point "need-to-know" checklist — owned network, one SLA, native AI, one bill, API plus console, one customer timeline — and applies that checklist side by side to an owned-carrier platform, a reseller platform, and a stitched multi-vendor stack.

**Why it matters** — Most "platform" labels hide a stitched stack underneath, and the seams surface after you've committed. The guide turns the evaluation into yes/no questions any vendor can be asked up front, and links into the [one provider vs multiple vendors](/vendors-comparison), [omnichannel comparison](/omnichannel-comparison), and [carrier-of-record](/compare/carrier-of-record) guides for the deeper cuts of each factor.

Public content pages depend on the orbit_content_pages table, and the
platform's post-startup self-heal task is what guarantees it exists. A
transient database connectivity blip during that task's catalog check
previously wedged the table's guaranteed-recovery permanently. The task now
treats an unreadable catalog as "assume absent" and still runs the
table-creation step safely, so public content page reads recover instead of
failing on every boot after a blip.

**What changed** — During sign up, picking a workspace logo showed a "ready" indicator, but the continue step navigated to the dashboard as soon as the upload resolved, so the success toast and saved checkmark never appeared. The workspace step now shows both — a success toast plus a checkmark next to the upload row — the moment your logo has been stored, before you're taken to the dashboard.

**What's new** — Teams whose workspace language is Spanish used to see a mixed-language Inbox: close hints, folder descriptions, macro editors, copilot menus, call cards, SLA banners, CRM context panels, and composer labels like "Ad attribution" or "Lifecycle" still rendered in English. The inbox now translates all of those into natural Spanish, matching what the rest of the dashboard already does.

**What changed** — The co-pilot card on Inbox conversations opened its Server-Sent-Events stream on a relative path. The browser resolves those against the dashboard origin, which has no API route, so the stream disconnected and the card stayed on "connecting" forever — no error, just silence. It now resolves through the same API-base helper every other streaming panel uses.

After a page reload, supervisors and admins see the live assist card update during the conversation, instead of a stall.

The CPaaS glossary (/glossary) now defines **number warm-up** — also called sender registration warm-up or 10DLC warm-up — the gradual traffic ramp on a newly registered sender rather than sending at full volume from day one.

The new page explains why carriers gate cold senders (a sudden burst with no delivery history reads as spam-like, so the volume gets filtered), how the warm-up differs from IP warmup (an email asset) and from brand vetting (The Campaign Registry's screening of the 10DLC brand), and what Orbit surfaces while a number warms: the warming phase (initial → ramp → steady → verified), today's daily_cap versus current_day_count, the 15-day progression forecast, and the 429 WARMING_QUOTA_EXCEEDED refusal that stops an over-cap send before it reaches the carrier. Pacing guidance is specific: split early volume across warming numbers or an already-warmed sender, and never retry-loop a refused send before the midnight-UTC cap reset. The definition frames warming as the carrier-side trust mechanism it is, with an on-page FAQ and related-term links to IP warmup, sender reputation, A2P 10DLC, and throttling.

German-language operators had a mixed-language contacts experience: the settings screens spoke German, but nearly the entire contacts surface — the list, the detail panel, the CSV import wizard, the merge-duplicates flow, segments, bulk actions, lifecycle stages, the AI memory view, and the GDPR right-to-erasure panel — showed English labels on controls and empty states alike.

That gap is closed. Choosing German in your account settings now renders the whole contacts area in German, and the English fallback only appears if a new label lands before its translation has been added. Genuine loanwords and product terms (Import, Tag, Segment, SMS, Audit) read the same in both locales, which is by design.