Public waitlist
Introducing Agent Loop on Vapi
Better voice agents, every call
Build self-improving agents on Vapi
You create an agent, tell it the goal, and it gets better on its own. Agent Loop runs the whole lifecycle for you: it builds the agent, scores every call against your goal, finds the misses, drafts a fix, tests it, and ships it to live traffic. Then it repeats.
- Simpler setup. Describe the job in a plain prompt, connect your tools and knowledge bases, and start taking calls. Composer picks the models, writes the prompts, and wires up the tools.
- Judged on your number. Set a goal in your own terms, like resolution rate or booked appointments, and score every call against it.
- Recursive self-improvement. The agent gets better with every cycle until it hits your goal, then keeps improving.
- Nothing hidden. Agent Loop runs as a harness on the existing Vapi platform. The new primitives it runs on are coming to the dashboard too, for even more control, not a black box.
Join the waitlist
Voice Agents that Improve Themselves
Setup
Composer builds toward a goal.
COMPOSER · GOALS
Observe
Clusters calls that didn't meet your goal into issues.
CALL HEALTH · ISSUES
Build
Draft an improvement based on the issues.
CALL HEALTH · IMPROVEMENTS
Test
Simulate calls based on the improvements.
SIMULATIONS
Deploy
Deploy changes as an experiment on real traffic.
TRAFFIC SPLITTING · EXPERIMENTS
Every part of Agent Loop also in the dashboard
Composer
Builds your agent's first version from a plain prompt, tools, and models.
Simulations
Hundreds of AI callers test every change before a real caller hears it.
Traffic Splitting
Send a slice of live calls to a new version, compare metrics, then roll it out.
Goals
One measurable pass/fail rule, in your business terms.
Call Health
Finds what went wrong, groups it into ranked issues, and drafts the fix to apply.
Experiments
Run versions side by side against a goal, and Vapi calls the winner and shifts traffic.