Word Is Bond
Start here

What it does, what a run looks like, how to run one

Three steps, in order. Read the first, open the second, follow the third. Nothing on this page needs an account until step three.

Try the judge Start free

1. What Word Is Bond does

An agent can stop doing its job without breaking: a model update, a platform change, or an edited prompt is enough, and nothing throws an error. Word Is Bond scores the conversation instead. A synthetic caller with a stated goal and persona talks to our hosted sandbox agent, or you send a transcript from one of your own agent's calls, and the judge returns a four-layer scorecard rather than one opinion: how the call flowed, whether the agent stayed on script, how it handled the caller, and whether the caller's goal was actually completed.

The point is the comparison. The judge is versioned and every scorecard names the judge that produced it, so a prompt edit, a model swap, or a changed integration shows up as a movement in a number instead of a report from an annoyed customer three weeks later.

Not ready yet: placing a test call to your own agent. On the free tier a direct target other than the sandbox returns a media link that your own harness must dial, and no harness is published yet; phone-network calls are on paid plans and no paying account has run one. This page will change when that path is proven on production.

2. What a scored run looks like

The scorecard is the artifact. It carries the pass or fail, the four layer scores with the judge's reasons, the measured metrics, and the transcript.

Run the judge on a real test transcript with no account, or open the illustrative sample report to see the full report layout.

3. How to run one yourself

The free tier needs no credit card and includes 60 test minutes a month over direct or SIP transport.

  1. Create an account and confirm your email address.
  2. Run one call against the shared sandbox agent and open its scorecard.
  3. Share the report with a link, or put the scenario on a weekly schedule.
  4. Create an API key in Settings to run the same check from a build.

With an API key, the sandbox run starts from the command line, and the exit status is what you wire into a build: 

# Run the sandbox scenario and fail the build if it did not pass
runId=$(curl -s -X POST https://api.wordis-bond.com/api/tests/run \
  -H "Authorization: Bearer $WORDIS_BOND_KEY" \
  -H "Content-Type: application/json" \
  -d '{"targetAgent":{"transport":"direct","peerId":"wb-demo-dental"},"goal":"Book a teeth cleaning and ask what it costs","persona":{"name":"First-time caller","description":"a new patient booking a routine cleaning"}}' \
  | jq -r .data.id)

curl -s https://api.wordis-bond.com/api/tests/$runId \
  -H "Authorization: Bearer $WORDIS_BOND_KEY" | jq -e '.data.passed == true'

The full walk-through, including scoring a transcript from your own calls, is in How to test a Vapi agent. Endpoint reference and scheduled runs are in the docs.