ما الجديد في Orbit

The global search API reference previously sketched the response envelope without a populated example, so there was no shape to code a palette or telemetry integration against. The page now opens with a complete round trip: a scoped two-pillar query and the 200 it returns, with results[] and by_pillar entries carrying id, title, href, and score, the per-pillar counts, and the partial array a client renders as a discreet per-pillar failure hint. A short note explains static_match_count — send the number of client-side shortcuts you rendered so a query that matched only static entries is never recorded as zero-result.

The integrations reference gains the Intercom connect flow on the same worked-sample pattern: start the OAuth install, hand the authorization URL to a workspace admin, read back the connected workspace and its subscribed topics, then confirm freshness with the status read-back and force a re-pull after a large import.

When a WhatsApp message is sent through a shared business account, Devotel Orbit quotes the cost up front and reconciles the wallet once Meta reports the final pricing of the conversation — refunding messages Meta marks free and adjusting the charge when the final category differs from the quoted one.

That reconciliation was failing on every eligible message before it could touch the wallet, so refunds and adjustments the reconciliation exists to perform never landed. The fix reads the charge actually recorded on the message and completes the refund or reprice from that figure; when no charge was recorded (the free-tier case), the wallet is left untouched instead of being credited an estimated amount.

The voice preferences reference previously showed its two worked API calls in cURL and TypeScript only. It now carries both calls in all six SDK languages, in the same tab order as the rest of the API reference, so you can copy the read or the write straight into your stack.

Each language uses its SDK's generic request escape hatch, so auth and the data / meta envelope behave exactly like the typed resources. The write path also shows the labelled 403 response a role-gated call returns, and the response examples now include the caller_scope field alongside the forwarding rule.

A persistent video room stays joinable indefinitely, but two closes can still land it in ended, and neither was named anywhere a customer would find them. The [video room lifecycle](/troubleshooting/video-room-lifecycle) runbook now opens a section on exactly that question, and the [troubleshooting hub](/reference/troubleshooting-hub) links it from its Video accordion.

- **What persistent means** — the room's expires_at is set years ahead, so the default 4-hour auto-end never fires, and when the room empties and closes, the next join re-opens it with its guest invites intact.
- **What it does not mean** — a live room that stays empty for ten minutes still closes (re-openably: the next join brings it back), and ending a room from the dashboard or API is final for every room, persistent included.
- **Reading an ended persistent room** — video.room.ended carries the answer: end_reason: room_finished is the re-openable empty-room close, while operator_end is an explicit End (or the 4-hour expiry path on a non-persistent room).

No behavior changed; this documents the shipped video room lifecycle.

The search box above the Recent Verifications table on the Verify page
used to match only the recipient, config name, channel, and status.
Pasting a verification ID — the one shown in a webhook payload, an API
response, or copied from a row — returned "No results", leaving you to
scan the list by eye.

The same search now also matches verification IDs, case-insensitively, in
both the table's search box and the filter bar above it. The placeholder
text now mentions ID, so what you can search for is visible at a glance.

When you open a verification in Recent Verifications, the Status timeline lists the events of its lifecycle. Until now, a verification that simply ran out of time left no footprint there: expiry was only recorded if someone submitted a code check after the window had already closed, so a verification that lapsed and was later cancelled looked as if it had been cancelled while still live.

The timeline now derives the expiry from the verification's validity window and records it as an Expired entry with the exact timestamp, positioned before any later event such as a cancellation. Verifications cancelled while their window was still open keep their existing history unchanged, and verified verifications continue to show Created and Verified as before.

On the Verify page, opening a verification whose validity window had
passed and pressing Resend code showed a "Verification has expired"
error, but the message list kept showing the old pending status until
the page was manually refreshed. The stored record was already marked
expired at that point — only the on-screen copy was stale.

The Verify list and the details drawer now refresh after any resend or
code-check error, so the status shown in the drawer, in the list, and in
the stored record always agree, with no refresh needed.

The menu item that opens the verification hub was labelled "Verify" —
the lone verb among otherwise noun-based labels like Inbox, Channels,
and Numbers. It now reads "Verification", in step with the rest of the
menu and with the "Recent Verifications" heading on the page itself.

The label appears translated in every supported language, and the
hub's address is unchanged, so bookmarks and saved links to it keep
working exactly as before.

US SMS compliance was documented control by control, with no single page that connected the pieces. The new [US SMS regulatory model](/concepts/us-sms-regulatory-model) concept page gives a compliance officer one read to answer three questions: what Devotel Orbit enforces, what you configure, and where the federal guard sits.

- **The three rule books, ordered** — TCPA sets the legal floor, the CTIA Messaging Principles turn it into the operational checklist carrier audits run, and carrier A2P registration and filtering enforce the result; the page maps each to the Orbit surfaces that answer it.
- **The opt-in and opt-out sums-of-parts** — the consent ledger, double opt-in, the consent gate, and the unknown-consent policies on one side; keyword rules, suppression lists, and the DNC and reassigned-number checks on the other, with each control linked to its owning page.
- **The quiet-hours envelope** — your per-channel settings (off by default), the pre-send policy scanner's lane gate, and the platform's single federal voice dialing-window guard, with the SMS-versus-voice asymmetry stated plainly.
- **Lane resolution** — why an untagged send is treated as marketing on the compliance surface: an exemption from a telemarketing rule has to be asserted, never assumed.
- **The health score as the read-out** — how consent coverage, STOP-reply rate, and carrier rejections close the feedback loop before they become throttling.

No behavior changed; this is documentation coverage for shipped compliance surfaces.

In Audience > Segments, the Tracked Link picker under “Retarget by link clicks” now opens when links are available and lets you select one. If there are no tracked links to show, it explains why and offers a retry instead of appearing unresponsive — and links that already loaded stay selectable even if refreshing the list momentarily fails.

The Concepts section gains a short explainer that summarizes how Devotel Orbit keeps customer data separate, for readers who want the shape of the guarantee in one sitting:

- A private database schema per organization, holding every working record you create, send, or receive.
- Server-side tenant resolution: each authenticated request is scoped to the organization behind your API key before any handler runs, and a client-supplied tenant identifier never redirects a request.
- A shared catalog limited to platform-wide routing and identity records, with an ownership check on every write.
- Per-record ownership gating for reseller families whose subaccounts share a schema by design.
- Fail-closed provisioning: a brand-new workspace answers with a brief 503 until its schema is ready, never a query against the wrong schema.

The page sits alongside the existing tenant isolation deep dive, which keeps the full engineering detail, and closes with pointers to data residency, the SOC 2 control text, and number ownership.

You can already author custom guardrail rules per agent, and the launch announcement said why the capability exists and where it sits. What it did not do is walk through the syntax at practitioner depth.

The new reference, [The Custom Guardrail DSL, Explained](/blog/tenant-guardrail-dsl-taxonomy-and-examples-2026), covers the rule surface field by field, the two trigger expression forms and how they compile, the evaluation legs and where your rules run in the turn pipeline, and the exact semantics of each action verb. It closes with five worked rules you can adapt to your own account: a no-refund promise guardrail, a no-medical-advice guardrail, keyword escalation tracking, a per-channel opt-out block, and a PII egress scrub, plus the line between rules you author and the platform's carrier-facing gates.

The capability itself is unchanged and remains live today. This is the buyer-side explainer practitioners were missing between the announcement and the effectiveness analytics guides.

Supervisors could already subscribe to real-time speech analytics for one live call — keyword flags as they are spoken, talk-to-listen ratio, dead air, interruptions, caller stress, and synthetic-voice risk — but there was no guide for building against the stream. The new page, [Supervisor live speech analytics stream](https://orbit.devotel.io/guides/supervisor-live-speech-analytics-stream), documents the endpoint contract (connected / status handshake, analytics events, the 30-minute reconnect window), which roles may open a stream and what a queue-scoped supervisor sees instead, where your keyword watchlist comes from, and how the same compliance redaction that masks spoken card numbers in the live transcript also masks them in keyword snippets. It closes with a runnable flag-wall web component that surfaces each signal as a chip and filters keyword chips by your category, and a troubleshooting pass for the common failure modes: an empty stream, missing keyword events, and the 403 a supervisor gets for a queue they are not assigned to. The page is listed in the voice navigation next to live transcription and in the supervisor section next to live monitoring and coaching.

## What's new

Two of the most common questions in voice AI now get a dedicated comparison on the blog:

- **Which engine hears your callers?** Whisper, Deepgram, and AssemblyAI compared on streaming posture, language coverage, and how each treats codec-degraded telephony audio.
- **Which house carries your brand voice?** ElevenLabs, Murf, and Cartesia compared on clone labs, studio workflow, and time-to-first-audio.
- **A decision tree, not a verdict.** Large-vocabulary telephony routes to the streaming specialists, clone-first brand voice routes to the clone labs, and everything else routes to Orbit's governed surfaces: the STT Playground, where your own clips decide the engine, and Model Presets, where the pick ships as a reviewed bundle.

The post deliberately prints no vendor latency or pricing numbers. Posture cells cite each vendor's public docs, and any number that decides your build gets measured on your own audio. [Read the head-to-head](/blog/stt-tts-engine-head-to-head-whisper-deepgram-elevenlabs-2026).

**What changed** — On the SMS page, the row of status filters above the
message table (All, Scheduled, Queued, Sent, Delivered, …) had no option
for received messages. Messages people texted to your numbers still
appeared in the list, but the only way to see them on their own was the
Inbound/Outbound direction toggle. There is now a Received filter that
shows exactly those messages, and the selection is kept in the page
address so a saved or shared link reopens the same filtered view.

Received messages were never missing from your account — everything
already delivered to your numbers is in the log exactly as before. This
only adds the filter that makes them quick to find.

The SMS rate card on the Billing and Billing → Pricing pages showed operator
names in the Country column for some destinations — Greece appeared as
"Cosmote - COSMOTE - Mobile Telecommunications S.A." and Algeria as
"Mobilis - Algérie Télécom" — with the same operator text repeated across
both columns.

Now:

- **Country** always shows the country name (Greece, Algeria, …), resolved
from authoritative country records rather than operator display strings.
- **Operator** shows just the operator, without repeating the country in
front of it.
- Country-wide default rates read as the country in **Country** and
"All operators" in **Operator**.
- Filtering by country now returns the full set of that country's operators,
including destinations that previously came back empty.

Rates, currencies, pricing-source labels and MCC/MNC codes are unchanged —
this is a display and filtering correction only.

## What's new

Teams evaluating email, lifecycle marketing, customer messaging, or in-app chat can now continue from a vendor comparison to a vendor-specific migration guide. The new runbooks cover SendGrid, Mailgun, Postmark, Braze, Intercom, CleverTap, and Sendbird.

Each guide lays out the records and workflows to account for, the Orbit-side mapping to verify, how to preserve tenant-owned preferences, and how to move in controlled stages with a rollback plan. The guides are explicit about where a provider's configuration must be rebuilt rather than assumed to import directly.

Every runbook links to its corresponding Devotel Orbit comparison and keeps the trade-offs in view, including cases where the existing provider remains a reasonable fit. [Read the migration runbooks](/blog).

**What changed** — When building segment conditions in Audience >
Segments, the Field menu now opens as a floating panel that positions
itself inside the visible screen. If the list of fields is longer than the
available space, the panel scrolls, and small arrows at the top and bottom
step through the remaining options one at a time.

**Why it matters** — Previously, opening the Field dropdown with many
available fields could push the list beyond the edge of the page, making
options hard to reach and breaking the layout of the segment form. Every
field is now reachable with the form staying put, on large and small
screens alike.

When you typed or pasted a very long value into the Value field of a segment condition, the condition preview (and the segment card on the Audiences page) grew with it, stretching the page and introducing unwanted scrolling.

Condition values are now capped at 256 characters — far more than any real email, phone number, tag or name needs — and longer pastes are rejected with a clear inline message. The same limit is enforced when conditions are saved, including values sent to the API from outside the dashboard, so an over-long value is rejected instead of stored. Condition previews on the segment dialog and on segment cards now truncate very long values with an ellipsis instead of expanding the layout.

When a channel is paused after unusual spend activity, sends and dials return the message "…is temporarily suspended by an automatic security pause after unusual spend activity." A new runbook, [Troubleshoot the automatic security pause](https://docs.orbit.devotel.io/troubleshooting/security-pause-suspended), explains what wrote the pause, how long it lasts, and how to lift it early from **Settings → Compliance → Active auto-mitigations** (owner role) or the DELETE /settings/compliance/fraud-mitigations/<channel> endpoint. It also shows how to tell the pause apart from a spend cap you configured and from an insufficient-balance refusal, and what to include when you escalate.

The page is filed under Reference → Troubleshooting, in the "Billing and account gates" section, and is linked from the troubleshooting hub next to the spend-cap runbook.