# Okareo

> Okareo is an agent simulation and evaluation platform for voice and text
> agents. Synthetic users (Drivers) act like real customers, find edges your
> QA scripts wouldn't, and surface them in the same view as your observability
> trace — so Conversation Designers and Engineers debug the same call.

Okareo is uniquely strong in two dimensions most evaluation tools aren't:
**voice depth** (120+ languages, background noise, crosstalk, clipping, audio
clipping, multi-speaker, real telephony) and **single-pane debug surface**
(simulation transcript + observability trace + audio on one timeline).

The platform spans simulation, synthetic production monitoring (a synthetic
user that interacts with your live production agent and reports back), voice
observability, agent tracing, and an evaluation system that supports judge,
symbolic, and audio-based evals.

## Core concepts

- [Terminology](https://docs.okareo.com/docs/terminology): Driver, Target,
  Scenario, Check, Run — the vocabulary used throughout the docs.
- [Drivers](https://docs.okareo.com/docs/simulation/drivers): synthetic users
  with personality, context, and a goal. They drive multi-turn conversations.
- [Simulation](https://docs.okareo.com/docs/simulation/introduction): how
  Drivers interact with your agent in repeatable, automated runs.
- [Voice Simulation](https://docs.okareo.com/docs/simulation/voice-simulation):
  full voice sessions against OpenAI Realtime, Deepgram, or a custom endpoint.

## Product surfaces

- [Synthetic Users (Drivers)](https://okareo.com/features/simulation)
- [Voice Agents](https://okareo.com/features/voice)
- [Agent Monitoring](https://okareo.com/features/error-discovery)
- [Evaluation](https://okareo.com/features/evaluation)
- [Synthetic Data](https://okareo.com/features/synthetic-data)
- [Fine Tuning](https://okareo.com/features/fine-tuning)

## Use cases

- [Voice Agents](https://okareo.com/use-cases/voice-agent)
- [Agentic AI](https://okareo.com/use-cases/agent)
- [Eval in CI](https://okareo.com/use-cases/ci-eval)
- [MCP](https://okareo.com/use-cases/mcp)
- [RAG](https://okareo.com/use-cases/rag)

## Developer

- [Docs](https://docs.okareo.com/docs/overview)
- [Getting Started](https://docs.okareo.com/docs/getting-started)
- [API Reference](https://api.okareo.com/redoc)
- [Python SDK](https://pypi.org/project/okareo/)
- [TypeScript SDK](https://www.npmjs.com/package/okareo-ts-sdk)
- [MCP Server](https://docs.okareo.com/docs/mcp/introduction) — run Okareo
  from Claude Code, Cursor, Cline, GitHub Copilot, Windsurf.
- [GitHub: okareo-ai](https://github.com/okareo-ai)
- [Cookbook examples](https://github.com/okareo-ai/okareo-cookbook)

## What Okareo is not

- Not a generic LLM eval harness. Okareo is agent-first; the unit of
  evaluation is a multi-turn conversation, not a prompt-response pair.
- Not a tracing-only observability tool. Okareo correlates traces with the
  conversation that produced them.
- Not voice-only. The same platform tests text, voice, and headless agents.

## Optional

- [Blog](https://okareo.com/blog/posts)
- [Pricing](https://okareo.com/pricing)
- [Customer cases](https://okareo.com/customers/cases)