Book demo

Get Started

Book demo

Get Started

Full Observability for custom voice stacks

Attach observers to the pipeline you already built. Your loop stays untouched, it just becomes measurable.

Get Started

Book demo

$ pip install tuner-core “tuner-stt-observer[deepgram]” “tuner-tts-observer[cartesia]” tuner-langchain

session = TunerSession(config=TunerConfig.from_env(), call_id=call_id)

session.attach(DeepgramAdapter(connection=dg_connection))

graph = session.attach(wrap_graph(your_graph))

tts = session.attach(CartesiaAdapter())

await session.flush()

Works with your providers

Deepgram

Cartesia

LangGraph

LangChain

Speechmatics

elevenlabs

Missing yours? Extend the base adapter, or send via the REST API

Setup

Integrate in under two minutes

Add a few lines to the voice agent you already have. Your audio and agent code stay the same.

Install

Pick the packages for your providers. You only install what you need.

Attach

Attach them to your existing STT, TTS, and LLM connections.

Run

Run the call like you always do. Tuner only records.

Flush

Call flush() at the end to send the timeline to Tuner.

Read the Pipecat guide →

Works with pieces you already chose

Each observer ships separately, install only what you use.

STT

Deepgram

Speechmatics

AssemblyAI

Coming soon

LLM / Orchestration

LangChain

LangGraph

TTS

Cartesia

OpenAI TTS

Coming soon

ElevenLabs

Coming soon

Anything else

Base adapter

Extend it for any provider

→

REST API

Send calls without an SDK

→

Why Tuner

See what's happening in production

Voice agents fail quietly, and at a scale no team can review by hand. Tuner turns every production call into structured data you can search, debug, evaluate, alert on, and test against.

Debug in minutes

When a call goes wrong, see exactly what happened and where — the full transcript, every turn, stage-level latency, tool calls, and conversation state. No more guessing from sparse logs.

Get alerted the moment something breaks

Don't wait for a customer complaint. Create alerts using red flags, metrics, evals, and multiple conditions, then get notified when a critical call fails or a problem starts appearing across your traffic.

Test before you ship

Run realistic call simulations over SIP before launch and after every change. Score simulations with the same evals you use in production, so regressions are caught before your callers find them.

Diagnose failures at scale

When you're handling thousands of calls a day, manual review doesn't scale. Tuner finds patterns across your traffic, traces failures to their source, and helps you turn what you learn into evals and alerts.

Measure what matters

Track outcomes, intents, extracted data, latency, cost, and custom evals across your Retell traffic. Filter down to the calls that matter and understand how your agent is performing over time.

Debug in minutes

When a call goes wrong, see exactly what happened and where — the full transcript, every turn, stage-level latency, tool calls, and conversation state. No more guessing from sparse logs.

Get alerted when something breaks

Measure what matters

Track outcomes, intents, extracted data, latency, cost, and custom evals across your Retell traffic. Filter down to the calls that matter and understand how your agent is performing over time.

Test before you ship

Run realistic call simulations over SIP before launch and after every change. Score simulations with the same evals you use in production, so regressions are caught before your callers find them.

Diagnose failures at scale

Everything you need to run your own custom voice agents in production

Turn production from a black box into something you can actually monitor, measure, and improve.

Catch failures early

Automatically flag hallucinations, broken flows, dead air, early hangups, and other failure conditions before they show up in your churn data.

See where latency comes from

Break out STT, TTS, and LLM latency at p50 and p90, so you can see exactly which part of the voice stack is slowing conversations down.

Get alerted when something breaks

Get notified when red flags, failed evals, or other conditions appear in production — while there’s still time to fix them.

Simulate calls before you ship

Stress-test your agent over SIP before launch and after every change, using the same evals that monitor your live traffic.

LangGraph & LangChain capture

Record LangGraph and LangChain node transitions, tool calls, and timing alongside session data, so you can see what your logic layer was doing during the call.

Track cost on every call

Attach a cost calculator and track LLM, TTS, and STT spend on every session — no separate billing pipeline required.

FAQ

Frequently asked questions

Common questions about connecting Tuner to your Pipecat agents.

Read the docs

Does attaching Tuner change how my pipeline runs?

Do I have to be on a particular framework?

Which providers are supported today?

What if I use a provider you don't list?

Do I have to install every observer package?

Does flush() block my call or add latency?

Can I capture LangGraph node transitions, not just the final output?

What if I built my own STT or TTS in-house?

Do custom stacks get the same evals, alerts, and simulation as platform integrations?

How is Tuner priced?

Is my call data private and secure?