Conversations & Analytics
Every call your agents handle is automatically recorded, transcribed, summarized, and scored — so nothing is lost and you can always see how well things are going. It all lives in your dashboard at your-infravoice-host, across three places: Call History (every call, with its full transcript and recording), QA (automatic scoring against rubrics you define), and Analytics (the big-picture dashboard).
No setup needed. Transcription, summaries, and metrics are captured for you on every voice test, phone call, and dialer call. You just review them.
Call History — every call, in one place
Call History (pinned near the top of the left sidebar) lists every voice session across all agents — browser Voice Tests, real phone calls, dialer calls, and campaign calls — newest first, updating live as calls come in.
The list
Across the top, a row of stat cards summarizes all your calls (not just the current page):
| Card | What it shows |
|---|---|
| Total Calls | Every voice session on record |
| Live Now | Calls in progress right now |
| Completed | Calls that finished normally |
| Failed | Calls that ended in error |
| Total Turns | Total back-and-forth exchanges |
| Avg Latency | Average agent response time |
| Total Time | Combined talk time |
Below that, filter and search to find any call:
- Agent dropdown — narrow to a single agent, or leave on All Agents.
- Status tabs — All, Live, Completed, Failed.
- Channel tabs — All Channels, WebSocket, Telephony, WebRTC.
- Search — by agent name, session, or status.
Each row shows the time, agent, status, channel (with the telephony provider and an inbound/outbound arrow for phone calls), turn count, duration, and average latency. Calls that moved between the AI and a human show a segments pill (e.g. 3 segments · AI → Human → AI), so a transferred call stays as one row instead of several.
Opening a call
Click any row to open the Call Transcript page for that call. There you'll find:
- Call Summary — a short, auto-generated recap of what happened on the call.
- QA Score — the call's score out of 100, if a rubric is attached to that agent (see QA below).
- Full Transcript — the whole conversation, turn by turn, with per-turn latency. For transferred calls it stitches every segment together in order (AI → human → AI).
- Recording — when a call was handled by a human agent, a Human Agent Recording player lets you listen back to that segment.
Tip: Cmd/Ctrl-click (or middle-click) a row to open a call in a new tab.
QA — score your calls automatically
Open QA (pinned near the top of the sidebar) to grade calls automatically against rubrics you define. It has two tabs: Score Dashboard and Rubrics.
Build a rubric
On the Rubrics tab, click New Rubric. A rubric is a set of weighted criteria the platform uses to grade each call:
- Give it a Rubric Name and an optional Description.
- Set a Pass Threshold (%) — the score a call must reach to pass.
- Add criteria — each with a name, a short evaluation guide, a type, and a weight:
- Yes / No — a simple pass/fail check (e.g. Did the agent greet the caller?).
- Scale 1–5 — a graded judgement (e.g. Was the agent empathetic?).
- Score 0–100 — a fine-grained score.
- Weights must add up to 100%.
In a hurry? Use Quick fill to drop in a ready-made Customer Service or Sales rubric, then tweak it.
Attach a rubric to agents
A rubric only scores calls once it's attached to an agent. On the Rubrics tab, click Bind on a rubric and check the agents it should grade. From then on, every call those agents handle is scored automatically.
One rubric per agent. An agent can have at most one rubric attached at a time — binding a new one replaces the old one for that agent.
Read the scores
The Score Dashboard tab shows scored calls with summary cards — Scored Calls, Avg Score, and Flagged — and a table of every scored call with its rubric, score, and result (Pass or Flagged). Filter by agent or tick Flagged only to focus on the calls that need attention. Click any row to expand the per-criterion breakdown, complete with the reasoning and a supporting quote from the transcript.
Summaries and QA use your Analysis LLM. Post-call summaries and scoring run on the optional Analysis LLM you can set on an agent's Providers tab. If you don't set one, the agent's main LLM is used.
Analytics — the big picture
Insights → Analytics turns all those calls into an at-a-glance dashboard. It refreshes on its own and shows a live badge whenever calls are in progress.
At the top, headline cards cover Total Calls, Total Messages, Total Duration, and Active Agents, plus an Active / Completed / Failed breakdown. Below that you get:
| Section | What it tells you |
|---|---|
| Pipeline Latency Breakdown | Where response time goes — ASR, LLM, tool execution, and TTS — with avg, p50, and p95, and a First Token total |
| Performance Summary | First-token latency percentiles and a per-agent breakdown of calls and speed |
| Channel Distribution | How calls split across WebSocket, Telephony, and WebRTC |
| Telephony & Direction | Inbound vs. outbound counts and a breakdown by provider (e.g. Twilio) |
| Tool Call Analytics | How often each tool was used, its average speed, and its success rate |
| Pipeline Health | Success and failure rates for ASR, LLM, and TTS, per provider/model |
Use it to spot slow tools, watch your latency budget, and catch provider errors before they add up.
Want alerts instead of watching? Insights → Alerts lets you create rules — for example, notify you when a provider starts erroring or when a call is flagged — so you don't have to keep the dashboard open.
In short: every call is captured for you. Browse them in Call History, open any one for its transcript, recording, summary, and QA score, define QA rubrics to grade calls automatically, and watch the whole operation from Analytics.