Skip to main content
Zelto is observability and benchmarking for AI agents, voice and text. It sits on top of your agent stack — Vapi, Retell, LiveKit, a chat or WhatsApp integration, or a custom setup — records every conversation, analyzes every transcript, and surfaces issues in how your agents handle them: mispronounced names, missed voicemails, conversations that drift from the script. Each issue is tied to the exact transcript moment it occurred, and you can benchmark prompt and agent versions against each other on the same monitors. Connecting a provider takes a few minutes and doesn’t change how your agents run. Once an issue surfaces, you can route it to Slack or file it in Linear.
The Zelto findings dashboard, showing recurring issues across an organization's agents with priority, conversation volume, and assignee.

How it fits together

Every interaction your agents handle becomes a conversation with a turn-by-turn transcript and, for voice, its recording. Analysis surfaces per-conversation observations; you promote the recurring ones into findings — durable, shareable issues — and record the fix as a change. For the whole pipeline end to end, see How Zelto works.

The core ideas

  • Agents — the AI agents (voice and text) and human agents in your organization.
  • Companies — the clients or brands your agents work for, with each conversation attributed to the right company.
  • Conversations — each conversation’s turn-by-turn transcript and recording (for voice), with on-demand AI summaries.
  • Findings — durable issues attached to the exact transcript moments that caused them.
  • Changes — agent-scoped fixes for one or more findings.
  • Experiments — compare real conversations from two prompt-version groups on the monitors that matter to a change.
  • Integrations — voice and text ingest, Slack and Google Chat notifications, Linear issue-from-finding, and open API / MCP access.

Works with your stack

Connect Vapi or Retell with an API key and a webhook URL you paste into the provider, forward finished sessions from a LiveKit worker, or POST conversations from any other stack — Bland, Pipecat, or your own. They all land in the same place, tagged with the originating agent. Not sure which path is yours? See Connect a voice provider.

Where to go next

Quickstart

Connect a provider and see your first transcript in five minutes.

Send traces

Export your first trace and identify agents with your own IDs.

How Zelto works

The whole pipeline end to end, from conversation to finding to change.

Guides

Triage findings, audit agents, and operate the platform.

API & MCP

Read everything programmatically over REST or MCP.