Skip to main content
Prerequisites: an active Tuner account. Sign up here if you haven’t already.
Creating an agent in Tuner is a guided flow. By the end you’ll have a call in Call Logs with its transcript, per-turn timing, and eval results — not just proof that data arrived, but proof that Tuner is judging it. Estimated time: 5 minutes

​
Step 1: Choose Your Provider

Click Add New Agent in the sidebar. The first thing you pick is your provider, and that determines the rest of the flow.
These platforms already hold your agent’s configuration, so Tuner reads it from them.
1

Add your platform API key

Paste in your Retell API key, or your Vapi Private key — not the public one. Tuner uses it to fetch your agents and to receive their calls.
2

Pick your agent from the list

Tuner loads your agents and pulls in each one’s details. Select the one you want to monitor.
No webhook setup needed — Tuner configures that side for you in the background.

​
Step 2: Pick a Template

Templates are what get you eval results on your very first call, rather than an unscored transcript. Each one pre-configures call outcomes, user intents, evals, red flags, alerts, and data-extraction fields for a common use case. Pick the closest match — it doesn’t have to be exact. A template is a starting point you shape, and everything it adds stays editable in Agent Settings.
The library changes over time, so what you see may differ from this list.
Prefer to define everything yourself? Configure via MCP lets you describe your agent to Claude, Cursor, or ChatGPT and have the Tuner MCP server generate the configuration. Configure manually starts from a blank slate.

​
Step 3: Add the SDK

On Retell or Vapi? Skip to Step 4 — nothing to install.
Tuner shows you a code snippet with your API key, workspace ID, and Agent Remote ID already filled in. Copy it into your agent, restart, and you’re capturing. That snippet is the minimal integration. Everything beyond it — cost calculation, recording URLs, disconnect reasons, agent versioning, SIP correlation for simulation — is on your integration guide:

LiveKit

Python and Node plugins for LiveKit Agents.

Pipecat

Observer SDK for Pipecat pipelines.

Dograh

The Tuner node inside your Dograh workflow.

Custom Stack

Python observers for your own STT, LLM, and TTS, or the API directly.

​
Step 4: Get Your First Call In

Two ways to finish, and the wizard offers both.

​
Take a Real Call

Tuner is already listening. Call your agent as you normally would, have a short conversation, and hang up — the call appears in Tuner within seconds of ending.

​
Or Run a Simulation

No call to make right now? Tuner can place one for you. The wizard has already configured the SIP connection, so the Run simulation button drops you straight into the simulation dialog — adjust the mix if you like, or just click Run. Tuner dials your agent with an AI caller, holds a conversation, and scores the result against the same evals as production traffic. See Call simulation for what you can configure.

​
What You Should See

Open Call Logs and click into the call:
  • Transcript — the conversation turn by turn, with tool calls inline
  • Eval results — pass, fail, or a score for each eval your template added
  • Voice metrics — per-turn latency, and where in the pipeline time went
  • Outcome and intent — how Tuner classified the call
If the call is there but nothing is scored, the template didn’t apply — check Agent Settings → Evaluation Rules. If the call never arrives, your integration guide’s troubleshooting section covers the usual causes.

​
What’s Next?

The Overview Dashboard

What the metrics, breakdowns, and red flag queue are telling you.

Call Simulation

Test your agent before launch and after every change.

Monitoring Routine

What to check, and how often, once calls are flowing.

Custom Evals

Add checks the template didn’t cover.