Signal vs Tuner

How Tuner works

Tuner turns production calls into structured data: intents, outcomes, extracted fields, eval scores and voice metrics. Rules combine these into labels and red flags, and alerts fire on a single critical failure or on a pattern.

To connect Retell, you save your API key and create an agent with the Retell agent id, and Tuner sets the webhook itself. Their docs say connecting “replaces the webhook URL on your Retell agent”; you re-add your old URL in their settings and Tuner forwards the original payload. Their docs do not describe a backfill of past calls.

Calls are classified into an outcome (success or failure, from your business’s point of view) and an intent (why the caller rang), kept as separate ideas. Both start from a library and take custom labels defined by a prompt. There are 21 predefined evals across accuracy, actions and tools, conversation quality, and safety and compliance, scored pass or fail or 1 to 5. Voice metrics are per turn, from provider timings.

Diagnose is an agent driven by Claude over Tuner’s MCP server. It reads your configuration, checks 30-day health, ranks red flags, isolates weak intent and outcome pairs, reads failing calls, and compares failures against the prompt. Then it can update prompts, create evals and add data fields.

Where Signal differs

Your webhook. Tuner takes over the Retell agent webhook and forwards to yours. Signal reads over the API and never touches it.

A board that persists. Tuner’s Diagnose explains failures when you ask, from the evals you set up. Signal keeps issues: each one placed on a step, grouped by the one change that fixes it, with a history across versions. If you reject one, it stays rejected.

Where the judgement lives. Several of Tuner’s evals match issues Signal finds by name: tool response mismatch, unnecessary repetition, dead air. Tuner’s are LLM judges. Signal labels what it measured apart from what a model judged, and says “cause not established” when nothing in the calls confirms a cause.

Proof. Signal compares versions on production calls, says whether a difference is real, and says “not enough calls” when that is the answer.

Where Tuner is stronger. Entry price, the number of stacks it connects, per-stage latency, a ready library of safety and compliance evals, PII redaction, and on-premises deployment for enterprise.

What it costs

Tuner bills in credits, about 5 a call by their own note, with a free tier and monthly plans that include a credit allowance. Evals do not seem to change the credits a call uses. Larger plans add PII redaction and a discount on overage.

At 10,000 three-minute calls a month, Tuner comes out lower than Signal; the price table on this page has both. At the low end, Tuner’s smallest paid plan costs less than Signal‘s and is sized for about 1,000 calls a month.

Signal charges a flat monthly plan with one balance of credits: each minute of a call costs a credit once it has started, and tests and the copilot draw on the same balance. Adding a check does not change the bill, and there is no charge per metric. There are no overage charges either: calls past the balance are stored, and analysed when the month resets or after an upgrade.

Using both

Tuner takes the Retell webhook and Signal reads over the API, so they run side by side.

1. Tuner’s evals and alerts tell you when something drops.
2. Signal tells you where in the flow it broke and which change would fix it, as an issue you can track.
3. After the change ships, Signal checks whether the issue shrank on production calls, and Tuner’s evals keep watching for the next drop.