github.com/combined-ai/isplay @v0.3.2

<img src="https://github.com/combined-ai/isplay/raw/v0.3.2/resources/brand/isplay-readme-logo.svg" alt="isplay" width="180">

Replay and analysis infrastructure for AI agents.

Website · Docs · Quick Start

Overview

isplay records agent runs as structured evidence, then lets humans or analyst agents test controlled counterfactuals against that evidence. It captures context, model calls, tool proposals, tool executions, artifacts, checkpoints, branches, replays, experiments, effects, and validity labels.

The goal is not to guess why an agent failed. The goal is to preserve enough evidence to replay, compare, and explain behavior with visible uncertainty.

What It Provides

• Structured run capture for agent systems.
• Checkpoints and branches for controlled interventions.
• Replay policies for recorded, live, or fixture-backed execution.
• Diff, metric, effect, and validity reporting for investigations.
• SDK, adapter kit, AI SDK integration, Mastra example, CLI, HTTP API, and docs app.

Quick Start

npm install
npm run build
npx isplay start

isplay start creates local artifacts, starts or reuses a Docker Postgres container, runs migrations, and starts the API server.

In another terminal:

export ISPLAY_API_URL=http://127.0.0.1:7373
npx isplay health
npx isplay projects create --name "Local Agent Lab"
export ISPLAY_PROJECT_ID="<id from JSON>"

Web Docs

The frontend and documentation app lives in packages/apps/web.

npm run dev:web
npm run build:web

Production deploys run from GitHub Actions through Vercel using the isplay project.

Repository Map

License

MIT