Catch what the agent says and does before your customer does.
Lens is end-to-end observability for voice agents on live speech-to-speech models. One screen shows the infrastructure and the agent together: who spoke when, every interruption and how the agent handled it, each stage the call moved through, every tool call, and how the call ended.
Before supervisors, sandboxes or simulation, we needed to see every call. You cannot fix what you cannot replay.
No 500 error when the agent talks over a caller or skips a step. Uptime stays green while the call goes wrong.
Lens first tells you which side broke, so the right person fixes it the first time.
What you can answer in 30 seconds
These are the questions that come up the morning after a pilot call goes wrong. Each one has a fixed place to look in Lens.
The same call data drives the fix loop.
Lens is not a dead end dashboard. The data a person reads here is the same data a coding agent reads through the CLI and MCP, and simulated calls land in the same view as real ones.
A blocked tool, an interruption, a stage that never advanced.
CLI and MCP expose the same transcript, events and verdict to a coding agent. CLI + MCP
The agent proposes a change to prompt, stage or tool policy, and ships it as a new version.
Simulated callers rerun the scenario. Results show up in Lens the same way. Simulation