Bolna gives teams a fast path to building and running production voice agents. Like any voice platform, a completed call is not the same thing as a good call: an agent can finish a conversation while giving an unsupported answer, missing a required field, or leaving the caller without a resolved outcome. VaaniEval imports Bolna agent executions into a review workspace so teams can inspect what actually happened on a call, not just that it happened.
What VaaniEval imports from Bolna
VaaniEval connects to a Bolna account with an API key and pulls the same execution data available through Bolna's own API, normalized into one workspace:
- Agent discovery. VaaniEval lists the Bolna agents on the connected account so a reviewer can choose which agent's calls to bring in.
- Conversation history. For a selected agent (or across all agents), VaaniEval fetches executions and supports custom date ranges, including one-click presets such as the last 7, 30, or 90 days.
- Transcript and outcome. Each execution's transcript, status, summary, and extracted fields are parsed into a normalized conversation record.
- Telephony and cost metadata. Call duration, hangup reason, answered-by-voicemail status, and cost breakdown are retained alongside the conversation for context.
- Call recording. Where a recording is available, VaaniEval fetches and proxies the audio so reviewers can listen to the call next to its transcript, without needing separate access to the Bolna API.
From call volume to call quality
A list of executions and their durations answers "how many calls happened" and "how long did they take." It does not answer whether the caller's underlying request was actually resolved. VaaniEval adds a reviewable scorecard on top of imported Bolna conversations: task completion, intent understanding, required-information capture, and delivery quality, each backed by a rationale that points to the specific part of the transcript that produced the score.
This distinction matters most in failure cases. A Bolna call can complete cleanly, with a polite agent and no dropped connection, while the caller's original task — rescheduling an appointment, resolving a billing question, confirming an order — never actually happened. Reviewing the transcript and recording together, rather than trusting call-completion status alone, is what surfaces that gap.
What is different about Bolna evidence
Bolna's execution detail returns a single formatted transcript string rather than a per-turn array with timestamps. VaaniEval parses that string into individual speaker turns for review, but because Bolna does not expose per-turn start and end times today, turn-level audio scrubbing is not available for Bolna calls the way it is for providers that return timestamped turns. Call-level audio playback and transcript review are still fully supported.
Bolna's recording URL also requires the same bearer-token authentication as the rest of its API, so it cannot be linked to directly from a browser. VaaniEval fetches the recording server-side and serves it back through the workspace, so reviewers get a working call player without needing direct API access themselves.
Setting up the workflow
Connecting Bolna follows the same pattern as other supported providers:
- Add a Bolna API key to the connected workspace.
- Select the Bolna agent (or agents) whose calls should be reviewed.
- Choose a date range and import the matching executions.
- Review transcripts, recordings, and metadata for each conversation.
- Run evaluation scoring and use the rationale and evidence to decide what to fix.
Because VaaniEval is self-hosted, the imported transcripts, recordings, and evaluation results stay in your own database rather than a third-party dashboard, and the same review workflow already used for other providers applies to Bolna calls without a separate tool.