GrowthOS · Real Agent Journeys and Evals

Description

GrowthOS is a platform designed to reconstruct and evaluate agent journeys in real-world scenarios. It focuses on understanding how agents discover, choose, and use products by reconstructing the task journey, which includes the agent's objective, the product surfaces encountered, and the evidence connecting these events. GrowthOS differentiates between production reconstruction and controlled Evals. While reconstruction describes real product behavior based on available evidence, Evals run defined tasks under controlled conditions to compare agents, models, environments, and product treatments. This approach allows for a comprehensive understanding of agent interactions and helps identify areas for improvement.

The platform requires data access depending on the task and surfaces involved, with a scoping call to identify the necessary evidence set. GrowthOS emphasizes the importance of evidence confidence, which qualifies the reconstruction process. It maintains separate metrics for production ATCR, Eval pass rate, and evidence confidence, ensuring that each measure answers different questions without collapsing them into a single score.

GrowthOS integrates with various product-owned evidence sources, including browser, docs, auth, API, MCP, CLI, SDK, and backend systems. It offers Python and TypeScript SDKs for instrumenting application and backend events. The platform's security measures include minimizing data before correlation, tenant boundaries, scoped identifiers, and pre-storage redaction.

GrowthOS is ideal for teams looking to improve their product's agent task completion rate by understanding real-world agent interactions and making data-driven improvements. It is trusted by teams at various companies and is led by a team of experienced founders with backgrounds in engineering and AI.

GrowthOS's Core Features

• Reconstruct agent journeys

• Run controlled Evals

• Measure production ATCR

• Integrate with product-owned evidence

• Python SDK for backend events

• TypeScript SDK for browser events

• Evidence confidence classification

• Security and privacy controls

How to use GrowthOS?

1. 1

   Configure: Set up data access and evidence paths

2. 2

   Use: Reconstruct agent journeys and run Evals

3. 3

   Optimise: Analyze results to improve product interactions

4. 4

   Verify: Measure real-world outcomes post-implementation

GrowthOS's Use Cases

• Agent journey reconstruction
• Controlled task evaluations
• Product improvement insights
• Evidence-based decision making
• Security and privacy management

FAQ from GrowthOS

• What is GrowthOS?

  GrowthOS is a platform that reconstructs agent journeys and evaluates tasks across agents and models to improve product interactions.

• How does GrowthOS reconstruct agent journeys?

  It reconstructs journeys by analyzing the agent's objective, product surfaces encountered, and evidence connecting events.

• What is the difference between reconstruction and Evals?

  Reconstruction describes real product behavior, while Evals run defined tasks under controlled conditions for comparison.

• What data access is required for GrowthOS?

  Data access depends on the task and surfaces involved, identified during a scoping call.

• How does GrowthOS ensure data security?

  GrowthOS uses tenant boundaries, scoped identifiers, and pre-storage redaction to ensure data security.

• What is evidence confidence in GrowthOS?

  Evidence confidence qualifies the reconstruction process by classifying connections from exact to unresolved.

• What integrations does GrowthOS offer?

  GrowthOS integrates with browser, docs, auth, API, MCP, CLI, SDK, and backend systems using Python and TypeScript SDKs.

• Who can benefit from using GrowthOS?

  Teams looking to improve their product's agent task completion rate by understanding real-world agent interactions.