Records

Overview

The Conversations page (also called “Records”) is where all your uploaded calls and chats live. From here you can upload new recordings, search and filter existing ones, view detailed analyses, and export data.

Records List

Client Admin Supervisor Agent

The main view shows a data grid with the following columns:

Column	Description
Participant	Name, phone, or email of the contact
Project	Which data source/project the record belongs to
Channel	Call or Chat
Duration	Length of the conversation
Quality Score	Mean of the six conversation metrics (0–100, one decimal), color-coded — the same score the dashboard averages. A record whose analysis rates none of the metrics has no score
Status	Current processing status
Conversation time	When the conversation happened, as reported by the source. A dash means the source reported no conversation time (file upload, some API submissions)
Received	When Aelo received the record. Hidden by default — enable it from the column gear. A gap between it and the conversation time shows delivery delay on the source side, not analysis time

Sorting: Click any column header to sort ascending/descending.

Pagination: Navigate pages at the bottom of the table, or change page size.

Record Statuses

Records move through a pipeline after upload:

Status	Meaning
Pending	Record uploaded, waiting to be processed
Transcribing	Audio is being converted to text
Analyzing	AI is evaluating the transcript
Delivering	Results being sent to CRM or other channels
Completed	All processing finished successfully
Failed	An error occurred during processing

Failed records can be retried using the Retry button in the record detail view.

Uploading Records

Client Admin Agent

1. Click the Upload button in the top-right corner.
2. In the upload modal:
   • Drag and drop audio files onto the drop zone, or click to browse.
   • Select the Project this record belongs to.
   • Choose the Channel type (Call or Chat).
   • Optionally provide participant details.
3. A cost estimate is shown before uploading (based on file duration and selected providers).
4. Click Upload to begin processing.

Supported formats: MP3, WAV, M4A, OGG, WEBM, FLAC

File limits: Maximum file size depends on your organization’s configuration.

Batch upload: You can select multiple files at once. Each file becomes a separate record in the selected project.

Project Status

Client Admin

Each project has a status, set from the project’s own settings page (Overview tab): Active, Paused, or Archived.

Status	Accepts new records	Appears in filters & analytics
Active	Yes — from the CRM integration, the public API, and file upload	Yes
Paused	No — every source is blocked (CRM integration, public API, file upload)	Yes
Archived	No	Dashboards: No · Records: Yes

Changing a project’s status never touches the records already in it — everything uploaded before the change stays in the Records list and in your analytics. An archived project drops out of the dashboard’s project filter, but keeps showing up in this page’s own Project filter, so you can still narrow the records list down to its history.

Filtering & Search

Use the filter bar above the records list to narrow results:

Filter	Type	Description
Participant	Text	Search by name, phone, or email
Project	Dropdown	Filter by project
Date Range	Date picker	Start and end date
Minimum Progress	Number (0–100)	Filter by minimum quality score (the list’s score column)
Agent	Dropdown	Filter by assigned agent
Status	Dropdown	Processing outcome — the same groups as the funnel on the Highlights tab: “Analyzed”, “Waiting to be started”, “Too short to analyze”, “No conversation recorded”, “In progress”, “Processing error”, “Other statuses”. Picking a group also shows missed calls, because the funnel counts them. “Delivery Failed” finds calls whose analysis is done but delivery to the CRM failed — the funnel counts them as analyzed
CRM Entity	Text	Filter by linked CRM entity

Text Search: with semantic search off, the search bar finds records by participant and agent name, phone numbers, line number, external ID, CRM activity name, and the beginning of the transcript. Letter case does not matter for Latin or Cyrillic letters — “oleg” finds “Oleg”, “олег” finds “Олег”. The exceptions are one-character queries and a rare fallback mode, most often for two-character queries: in both, only the basic Latin letters A–Z ignore case. If a short query with other letters (Cyrillic, accented) misses a record, type at least three characters. When a query matches a very large number of records, a note above the results says “At least N matches — the count is not final”: the list keeps loading as you scroll, but the total shown is a lower bound. Refine the query or narrow the period to get an exact count. In the web app, when many other records match first, search keeps going through the matches — “Searching further through the matches…” — and after a few steps offers a Search further button to continue. If part of the results came from the fallback mode, a note above the list says some records may be missing: try again in a minute, or type at least three characters. The same search works on the Records tab of the Bitrix24 app, with the same notes and the same Search further button: results load 20 at a time, and Load More adds the next ones.

Semantic Search: Use the search bar for natural language queries across transcripts (e.g., “customer complained about delivery”). The filters you have selected — status, project, period, agent and the rest — apply to the results too, and the number next to the page title counts what was found: up to the 10 closest matches. If the filters set some of the closest matches aside, a note above the results says there may be more — make the query more specific to find them.

Saved Filters: Save frequently used filter combinations for quick access.

Deep linking: Filters are reflected in the URL, so you can bookmark or share filtered views.

Record Detail Panel

Click any record in the list to open the detail panel (slide-out sheet).

At the top is a header summarising the record: quality score, conversation type and duration, line number, category, deal outcome, CRM entity link, position in the call series, project, conversation time and the time Aelo received it, sale probability.

Below it the panel is split into tabs: Transcript, Notes and — only in a narrow panel (under 900 px) — Score. In a wide panel the analysis is always visible to the right of the transcript, so it needs no tab of its own.

AI assistant

On the Pro plan and above, the panel header (and the app sidebar while a record is open) has an Assistant button. It opens a side panel where you ask questions about this record — pick one of six suggested prompts or type your own (up to 2,000 characters). The answer is grounded in the record’s transcript and analysis: a thesis carries the quote it rests on whenever the assistant found one in this record, and a quote it matched to one place in the transcript shows a ▶ mm:ss timecode — clicking it plays that moment (for chats, it scrolls the transcript there). A quote it could not match is marked “Not confirmed in the transcript”, and one that turns up in more than one place is marked “Found in several places”; neither carries a timecode, because there is no single moment to play. A thesis the assistant could not back with this record stands without a quote. The panel always names the record it is answering about, right under its title. You can keep browsing with it open — it is not a modal window — and it moves to whichever record you open next, together with that record’s own conversation and its own unsent draft. While an answer is still being prepared it stays where it is, so nothing you have paid for is lost; it then says the record is “not on screen” and offers to move over when you want to. On a screen without a record — a dashboard, say — it keeps the last one and says so rather than going blank. Each organization has a daily question limit; the panel shows how many questions are left today. Conversations stay in the panel for later reading and are deleted when the record is deleted — usually together with it, otherwise by a routine cleanup.

Asking about a deal, lead, contact or company. When the open record is linked to a deal, lead, contact or company in your CRM, the record header shows Ask about the deal (or the lead, contact or company), and the panel gets a scope switcher: This record or This deal. A question about the deal is answered from all of its calls and chats that you are allowed to see — conversations you cannot open are left out — and a quote from another call or chat is marked as such and opens that conversation when clicked. For a long history the assistant reads the conversations most relevant to your question in full and the rest as a one-line summary each, and says so when it could not read everything. Conversations about a deal are kept per deal, the same way as for a record. A question and its answer are deleted when any record the answer was built from is deleted — usually together with it, otherwise by a routine cleanup — and in any case at the end of your organization’s retention period. The daily question limit is shared by both scopes.

Asking about all your conversations. The AI assistant item in the app menu opens a page for questions about every call and chat you are allowed to see — conversations you cannot open are never used. In the web app the panel’s Whole base button and the Assistant button on a screen without a record open this page too; a question you typed in the panel stays there, with the record it was typed for. Inside Bitrix24 there is no such page: there Whole base and the Assistant button ask about the whole base right in the panel, without filters. Narrow a question by period, project, agent, type (calls or chats), category or customer; the customer field suggests names as they are stored in your records, and a name is picked from the suggestions. Only customers with conversations under the other filters are suggested; change another filter after picking, and the customer is removed with a note — the name stays in the field, and the suggestions offer it again if it still fits. If nothing matches all the filters together, the answer says so and suggests removing one. The assistant finds the conversations that match your question by meaning, reads up to 12 of the closest in full and answers from them, with quotes that open the conversation they come from; the answer says how many records it searched and which filters it was asked under. If nothing matches, it says so without asking the model. Such an answer takes longer than one about a single record — it can take a few minutes — and is prepared in the background: you can leave the page and come back to it. While it is on its way the filters and the list of conversations stay closed. Changing the filters starts a new conversation, so an answer never mixes history asked under other filters; the page lists your latest five whole-base conversations. Only conversations that have been analysed and are already in the search index are used — one still waiting for analysis, or whose analysis failed, is left out. When you see only part of the base — as an agent or a team lead — or filter by customer or category, the assistant searches the 5,000 most recent analysed conversations that match. The daily question limit is shared with the other scopes, and a question and its answer are deleted when any record the answer was built from is deleted, and in any case at the end of your organization’s retention period.

Audio Player (Calls)

• Waveform visualization — visual representation of the audio
• Playback controls — play, pause, skip, speed adjustment
• Speaker labels — color-coded segments by speaker

Transcript Tab

• Full conversation transcript with speaker labels and speaker diarization
• Word-level timestamps for each segment
• Click a transcript segment to jump to that point in the audio
• Keyword highlighting when search terms are active

Analysis

In a wide panel the analysis occupies the right column next to the transcript; in a narrow one it opens on the Score tab. It shows:

• Quality score with breakdown by category
• Sale probability
• Sentiment indicators
• Key moments and topics
• Compliance flags
• Recommendations (see Quality Analysis for details)

Inside the analysis there is a group filter: All, Score, Script rules, Speech, Insights. A group’s button appears only when that group has data, and the filter itself only when at least two groups do.

Script rules group

If the project has a linked sales script and adherence has been evaluated, the analysis gains a Script rules group holding the “Script Adherence” card and its score. It is a group inside the analysis, not a tab of the panel. See Scripts for details.

Notes Tab

Notes and annotations on the record; the tab label carries their count. See the Annotations section below for how to add them.

Record details and processing status

Project, participant, channel, duration and the CRM entity link live in the panel header — there is no separate Metadata tab. The processing stages are listed above under “Record Statuses”; the panel has no pipeline-status tab either, but when processing has been halted the panel shows a notice with the reason.

Annotations

Client Admin Supervisor

Add notes and annotations to specific parts of the transcript:

1. Select a transcript segment.
2. Click Add Annotation.
3. Type your note and save.

Annotations are visible to all users who can access the record.

CRM Entity Linking

If a CRM integration is configured, records may show linked CRM entities (contacts, deals, leads). Click the CRM link to navigate directly to the entity in your CRM.

Bulk Actions

Client Admin

Select multiple records using the checkboxes to perform bulk actions:

• Delete — permanently remove selected records
• Archive — move records to archive
• Re-analyze — resubmit selected records for AI analysis

A confirmation dialog appears before destructive actions.

Export

Client Admin Supervisor

Export records data in multiple formats:

• CSV — spreadsheet-compatible data export
• PDF — formatted report for sharing
• Markdown — text-based export

Click the Export button above the records list. The export includes all records matching current filters.

Re-analyze & Retry

• Re-analyze (Client Admin Supervisor): Resubmit a completed record for fresh analysis (useful after changing evaluation profiles).
• Retry (Client Admin Supervisor): Retry a failed record that encountered a processing error.

Both actions are available from the record detail panel.

In manual mode, Retry runs nothing: the record waits to be started, like any other record waiting for a start, and no minutes are spent until it is started. The same happens when Auto-Analysis is off in the record’s project and it was the analysis that failed, or the record already had an analysis (it was started by hand); a failed transcription or delivery of a record without one is run again the way the project is set up, without an analysis. Only a client or admin can start a record, so where the record would have to wait, a supervisor’s retry is refused. A record whose processing was already paid for is not charged again when started, and the price shown before the start leaves it out. A call whose transcript is saved but whose minutes were never charged is charged once when a start, a retry, a delivery retry or a re-analysis analyses that saved transcript — for the length of the conversation, or for the length of the recording when the conversation was not measured. When neither length is known, a start, a retry of a failed record or a delivery retry does not analyse the saved transcript: the call waits for its recording, like a call that arrived without one. A re-analysis of such a call, whose length is not known, is refused: nothing is analysed and nothing is charged. When the source answers that it has no recording of a call whose transcript is saved but not charged, a start analyses that transcript and charges it instead of closing the call.

Access by Role

Feature	Client	Admin	Supervisor	Agent
View records list	✅	✅	✅ (project-scoped)	✅ (own records)
Upload records	✅	✅	❌	✅
View record detail	✅	✅	✅ (project-scoped)	✅ (own records)
Add annotations	✅	✅	✅	❌
Bulk actions	✅	✅	❌	❌
Export	✅	✅	✅	❌
Re-analyze / Retry	✅	✅	✅ (project-scoped)	❌
Delete records	✅	✅	❌	❌

Tips & Best Practices

• Use Saved Filters to quickly switch between common views (e.g., “Failed records this week”).
• After uploading, check the record status. Most records complete within 2–5 minutes depending on duration.
• Use Re-analyze after updating your evaluation profiles to see how scores change.
• Supervisors: filter by agent to review individual team member performance.
• Agents: your own records are automatically filtered when you access the Conversations page.