Skip to main content

Changelog

Stay up to date with InsightAgent releases and improvements.

May 2026​

Billing & Usage redesign​

The Billing & Usage section is rebuilt into three focused views, each surfaced as its own sub-nav item in the sidebar:

  • Current period (/billing) — live usage and what your current cycle is costing you. New side-by-side layout: progress + pacing on the left (Remaining / Days left / Projected), itemized Billing Breakdown on the right (plan rate, included minutes, each add-on listed separately, projected overage with the date it will bill).
  • Past periods (/billing/history) — closed cycles with daily breakdowns, per-interview detail, and CSV export. New: each row shows a derived status (Paid / Partially paid / Unpaid / Overdue / Pending) based on the invoices that touched it. Drilling in shows an Adjustments section when credits or discounts have been applied to that cycle.
  • Invoices (/billing/invoices) — the new invoices ledger. Click any row to expand into a per-line-item breakdown grouped by the period each line covers.

Combined statement model

Each invoice now carries line items from up to two cycles: next cycle's base + add-ons (billed in advance) plus the just-closed cycle's overage (billed in arrears). The current-period Billing Breakdown labels this explicitly — your base shows "billed [date] ✓ (Invoice #N)" pointing back at the prior invoice, and the upcoming overage line shows "projected · bills [date]" pointing forward. The "Period total" footer reads "across 2 invoices" so the split is clear.

Next payment card

A forward-looking preview of the upcoming invoice with both advance and arrears lines visible. Replaces the prior "Estimated Total" framing — the date and the breakdown make it unambiguous what bills next and when.

Scheduled plan changes

When a plan change is queued, a banner appears at the top of all Billing pages: "Starting [date], your plan changes to [tier]. This period continues on [current tier]." Plan changes always take effect at the next billing cycle boundary — never mid-period. The Next payment card recomputes immediately to show what the upcoming invoice will look like under the new plan. Overage for the current cycle still bills at the rates that were in effect when the usage happened.

Adjustments (credits, discounts, compensations)

Credits attached to an invoice (e.g. an SLA service credit for downtime) now flow into the Adjustments section of the affected cycle's drill-down on Past periods. The cycle's Billing Breakdown shows the original total, the adjustments, and the net side-by-side, with a link to the invoice that carried the credit.

See Billing & Usage.

Phone Interview Completion​

Parallel to the Interview Completion change for web sessions: phone interviews that reach a natural conclusion are now marked Completed automatically, whether or not the expert explicitly signals completion. The system reviews the conversation transcript after each call and upgrades the classification if the conversation actually wrapped up cleanly.

In addition, interviews that stay In Progress past 24 hours of inactivity are now automatically closed out as Complete — at that point the cycle has clearly ended even if no clean completion signal was received. This applies across all interview types (phone, web, copilot). Analysis generation and post-interview webhooks fire on close, so your downstream integrations receive the same data they would from a natural completion.

Expect a corresponding shift in the Billing & Usage outcome split — some sessions that would previously have stayed in the Incomplete bucket will now appear as Completed.

Multi-Agent Routing​

A single call can now be handled by more than one agent. A main agent transfers the live conversation to a sub-agent partway through — same voice, no re-introduction, no break the expert can hear.

Agent roles

  • Agents are now Main (attachable to interviews, can transfer) or Sub-agent (reachable only by transfer). Settings → Agents has Main / Sub-agents / All tabs.

The agent chain

  • Define transfer rules on a main agent: pick a target sub-agent and describe, in plain English, when the handoff should fire. The agent decides in real time from the live conversation.
  • Optional per-rule delay, audible transfer message (off by default), and target re-introduction (off by default).

Seamless-handoff safeguards

  • Voice-mismatch and persona-name-mismatch warnings appear on any rule that would make the interviewer's voice or name change mid-call.
  • "Copy persona from existing agent" clones voice, persona name, prompt, and first message so a matched sub-agent is one click away.

After the call

  • The interview detail page shows a one-line handoff summary (the agent chain, with timestamps and the condition that fired on hover) and inline handoff markers in the transcript. Markers are excluded from exports.
  • Each sub-agent's page lists its inbound rules — every main agent that can transfer into it.

See Multi-Agent Routing.

Interview Completion​

This change applies to all web interviews, not only multi-agent ones.

Previously, a web interview was marked Completed only when the expert clicked "I'm done" at the end. If the session ended after the conversation had naturally wrapped up — without that click — it was recorded as incomplete, even though the interview was effectively finished.

From this release, web interviews that reach a natural conclusion are marked Completed automatically, whether or not the expert clicks "I'm done". Sessions that drop mid-conversation are still treated as incomplete and remain resumable exactly as before — the expert can rejoin and continue.

This applies going forward as sessions are processed; it does not retroactively re-label interviews from before the release. Expect a corresponding shift in the Billing & Usage outcome split — some sessions that would previously have fallen into the Incomplete bucket will now appear as Completed.

Agent Voices​

You can now choose the voice an agent speaks with, from a curated list filtered to the agent's language. The same voice applies to both the Phone and Web versions. Agents created before this change keep their existing voice until you pick a new one.

See Agent Voices.

Usage Dashboard Uplift​

The Billing & Usage page now provides real-time visibility into how minutes are being consumed throughout the period, with per-interview detail and CSV export for offline analysis.

Daily breakdown for the current period

  • New bar chart showing billable minutes per day, with a reference line for your tier's included minutes.
  • Cumulative month-to-date overlay so you can see at a glance whether usage is trending toward your included allotment.

Per-interview detail table

  • Every billable interview in the current period is listed with: interview ID, expert name, title, completion time, session count, billable minutes, and end reason.
  • Expand any row to see the per-session breakdown (start/end times, duration, billable minutes, and the disconnect reason for that specific session).

Outcome classification

  • Each interview is automatically classified into one of three buckets:
    • Completed (single session) — finished cleanly on the first attempt.
    • Completed (with retry) — the expert reconnected one or more times before finishing successfully. Still billed as a normal completion.
    • Incomplete — the final session ended without a clean completion (network drop, expert disconnect, silence timeout, etc.).
  • A summary card shows the count and percentage split across the three buckets, making it easy to spot a spike in incomplete sessions.

CSV export

  • "Export CSV" button downloads the current per-interview table for joining against your own data warehouse or BI tools.
  • Column order is stable (interview_id, expert_name, title, final_completed_at, session_count, billable_minutes, final_disconnect_reason, classification) and titles containing commas or quotes are properly escaped.

The existing period totals, usage history, and tier information remain unchanged.

See Billing & Usage for details.

Incremental sync filter on GET /api/interviews​

Two new query parameters land on the interview list endpoint:

  • completedAfter (ISO 8601 timestamp) — returns only interviews with completedAt strictly after the value. Designed for incremental polling: store the completedAt of the last interview you ingested and pass it back next time to fetch only the delta.
  • sortOrder (asc or desc) — direction for the completion-time sort. Default asc (oldest-completion-first) matches the iteration pattern most polling clients expect.

Pair with status=COMPLETED for the typical case:

GET /api/interviews?status=COMPLETED&completedAfter=2026-04-28T12:00:00Z

No changes to existing behavior — both parameters are optional, the default sort (createdAt DESC) is unchanged when completedAfter is absent. See List Interviews for the full parameter table and the incremental-sync pattern.

New disconnectReason value: user_audio_silent​

Web sessions that end without any detectable expert speech are now labelled with a dedicated disconnectReason of user_audio_silent. This is typically caused by a microphone or device problem on the expert's side — a wrong default input device, muted hardware, or a broken capture pipeline — where the expert may not realise their setup isn't producing any audio.

Previously these sessions were labelled as the generic voice_session_disconnect, making them indistinguishable from sessions that had real audio but ended unexpectedly. Filtering on disconnectReason=user_audio_silent now lets you identify and aggregate silent-failure sessions in your own pipeline.

No API contract changes — disconnectReason is a string field that already accepts arbitrary values; the new value flows through GET /api/interviews/:id without any client changes required. See Session disconnectReason values for the full vocabulary.

April 2026​

Reliability Improvements for Expert Join Flow​

A set of fixes to make the expert interview join flow more robust against browser-side connection failures (microphone permission denied, transient network issues, etc.).

Stricter "started" semantics on the join page

  • The expert interface now only treats an interview as "started" once the conversation session is actually live, not on the Start button click.
  • If microphone access fails or the connection can't establish, the expert stays on the Start screen with a clear error and can retry — no more advancing into a partially-broken state.

Server-side guard against ghost-completion

  • PUT /api/ungated/interviews/:id with status=COMPLETED is now refused with 409 NO_SESSION_TO_COMPLETE when the interview has no captured session. Prevents interviews from being marked complete with no audio or transcript.

Clear empty state for completed interviews with no recording

  • The expert join page and the dashboard interview view now render a "No conversation was captured" message when an interview was marked complete without any recorded audio — instead of an indefinite "Processing" spinner.

These changes are non-breaking for existing API integrations. The new 409 response code on the ungated endpoint is only returned in the edge case where a completion is attempted on an interview with no captured session — a state that should not occur in normal use after these fixes.

February 2026​

API Key Management​

Create and manage your own API keys directly from the InsightAgent dashboard.

  • Self-service key management via Settings > API Keys
  • Create, rename, and revoke keys instantly
  • Up to 5 active keys per account
  • Keys displayed in masked format after creation for security
  • Revoked keys are rejected immediately

API endpoints — four new endpoints for programmatic key management: list, create, rename, and revoke. See API Keys for details.

Custom Vocabulary for Speech Recognition​

Improve transcription accuracy with domain-specific terms for company names, technical jargon, and acronyms.

Account-Level Vocabulary Management

  • Organize terms into categories (e.g., "Pharma Companies", "Drug Names")
  • AI-powered term suggestions based on interview context
  • CSV import/export for bulk vocabulary management
  • Up to 1,000 terms across 20 categories

Interview-Specific Overrides

  • Select or deselect terms from your account vocabulary per interview
  • Add custom terms that only apply to a single interview
  • Changes saved automatically

Automatic Vocabulary Injection

  • System auto-generates terms from interview context (questions, expert data) at call start
  • Merges with account vocabulary and applies interview overrides
  • Injects into speech recognition for real-time accuracy improvements

AI Transcript Cleanup

  • Post-interview AI review corrects remaining speech recognition errors using vocabulary context
  • Fixes common ASR mistakes like misspelled company names, drug names, and technical terms

See Custom Vocabulary for full documentation.

Minor Improvements​

  • Analysis now generates automatically for auto-completed calls
  • Flexible employment history parsing for expert profiles
  • Dial-in fields no longer revert when polling refreshes the page
  • Meeting ID auto-strips spaces for easier paste input

Human Detection for Phone Conferences​

Smarter handling of phone interviews when the expert hasn't joined the conference yet.

Intelligent Waiting

  • AI agent now waits silently after joining a conference bridge
  • Uses AI-powered speech classification to detect when a real human speaks
  • Ignores hold music, system prompts, and automated announcements
  • Interview begins automatically when the expert starts talking

New Status: Waiting for Expert

  • Amber status badge with animated pulsing indicator
  • Clear banner showing "Agent connected to conference — waiting for expert to join"
  • Interview locked during waiting state (cannot edit)

Timeout Protection

  • 5-minute timeout if no human is detected
  • Call ends automatically with "no show" tracking
  • Prevents indefinite waiting on empty conferences

See Conducting Interviews for details.

January 2026​

Live Transcript​

Monitor your interviews in real-time with live transcript streaming.

Real-Time Observation

  • Watch the transcript populate as the conversation happens
  • See both agent and expert messages appear instantly
  • No need to wait for the interview to complete

Session Tracking

  • Clear visual indicators for active vs. inactive sessions
  • Session separators show when multiple sessions occur
  • Abandoned session tracking helps identify dropped interviews

Interview List Improvements

  • New status filters to quickly find interviews by state
  • Search functionality to find interviews by expert name or topic
  • Sticky header and footer for better navigation
  • Abandoned interview indicators in the list view

Analysis Enhancements

  • Key quotes now include only expert quotes (agent quotes filtered out)
  • Improved quote extraction accuracy

Custom Variables for Agent Prompts​

Personalize your AI agent prompts with custom variables that can be filled in differently for each interview.

Variable Management

  • Create reusable variables with display names and descriptions
  • Mark variables as required or optional
  • Set default values for optional variables
  • Variables are shared across all agents in your account

Interview Integration

  • Custom Variables section appears when creating/editing interviews
  • Required fields marked with red asterisk (*)
  • Default values pre-filled automatically
  • Validation prevents saving without required values

Agent Workflow

  • Reference variables in prompts using {{variable_name}} syntax
  • Preview variables when changing agents on existing interviews
  • Variables are substituted at runtime during the interview

Settings Page

  • New Settings → Variables page for managing account variables
  • See which agents use each variable
  • Protected deletion (can't delete variables in use)

See Custom Variables for full documentation.

AI-Powered Summary & Quotes​

InsightAgent now automatically generates executive summaries and extracts key quotes from your completed interviews.

Summary Generation

  • Concise executive summary of each interview, highlighting main topics and key insights
  • Factual analysis based only on what was actually discussed
  • Available immediately after interview completion

Quote Extraction

  • Notable verbatim quotes extracted from the transcript
  • Each quote includes speaker attribution and context
  • Clickable timestamps to jump directly to that moment in the recording
  • Strict validation ensures only exact quotes from the transcript are included

Account Controls

  • Toggle summary generation and quote extraction independently
  • Settings apply to all new interviews in your account
  • Access via Settings → Post-Interview Analysis

Manual Regeneration

  • Regenerate analysis anytime from the Analysis tab
  • Useful when switching between session recordings

See Interview Analysis for full details.

Minor Improvements​

  • Quotes are now sorted chronologically by timestamp
  • Activity log now tracks when analysis is generated or regenerated
  • Session switching preserves audio player state
  • Improved interview completion tracking in activity log

December 2025​

Latest Updates​

Check back here for the latest product updates and new features.

Release Notes Format​

Each release includes:

  • New Features - Brand new capabilities
  • Improvements - Enhancements to existing features
  • Bug Fixes - Issues resolved
  • Breaking Changes - Changes that may require action

Stay Informed​

  • Subscribe to release notifications in your account settings
  • Follow us on LinkedIn
  • Check this page regularly for updates

Feedback​

Have suggestions for new features? Contact us at feedback@insightagent.io