Skip to main content
Open source

Test AI agents the way people talk to them

IBM VIBE is a conversation-centric testing suite. Script realistic conversations, run them against your agents, and inspect transcripts, token usage, similarity scores and failures.

Get startedProduct tourGitHub

Run a conversation from the CLI

vibe conversation create --name "Refund flow"
vibe conversation add-message --conversation <id> \
  --role user --content "I want a refund"
vibe conversation set-target --conversation <id> \
  --target "I can help with that" --threshold 65
vibe run conversation --conversation <id> --agent <agent_id>

Why VIBE?

Test conversations, not just prompts

Script the multi-turn exchanges your users actually have, with expected replies per turn.

Make failures inspectable

Review sessions and transcripts with timing, token usage and similarity scores instead of a black box.

Compare agent versions

Iterate on prompts, tools and LLM settings against the same evaluation inputs.

Quick execute

Run one conversation against one agent in seconds, then scale up to suites.

CLI and REST API

Script everything with the vibe CLI or the REST API, with JSON output and CI-friendly exit codes.

Local and reproducible

Runs on your machine with SQLite-backed storage. No cloud service required.

A conversation-first workflow

  1. 1Configure
  2. 2Script
  3. 3Execute
  4. 4Inspect
  5. 5Iterate

See it in action

VIBE dashboard with first-run checklist and summary metrics
Dashboard
Conversation detail page with the script and latest session
Conversations
Session page with turn-by-turn messages and similarity scores
Sessions

Start in two minutes

git clone https://github.com/IBM/VIBE.git && cd VIBE
npm install
cp backend/.env.example backend/.env
cp frontend/.env.example frontend/.env.local
cp agent-service-api/.env.example agent-service-api/.env
npm run dev

Then open http://localhost:3000. Read the quickstart for the first-run path, or follow news and tutorials on the blog.