isplay Documentation
Replayable investigation infrastructure for AI agents, with structured capture, counterfactual replay, fixture-aware experiments, and bounded evidence reports.
isplay captures agent runs as structured evidence and lets humans or analyst agents ask controlled counterfactual questions about that evidence. It records what entered context, what the model proposed, what tools actually did, which checkpoints are safe to branch from, what changed under replay, and how trustworthy each conclusion is.
Core idea
isplay is not a root-cause oracle. It is an evidence system: every conclusion should cite the captured trace, the replay policy, fixture provenance, diffs, metrics, effects, and validity labels that support it.
Install
Use the skill, or choose an adapter and wire capture manually.
Quickstart
Capture a run after installation and start the first investigation.
Tutorials
End-to-end learning paths for first capture, first replay, hypothesis batches, and the Mastra demo.
Concepts
Standalone explanations for runs, events, context, branches, replays, fixtures, experiments, validity, and storage.
Guides
Task-oriented workflows for instrumentation, replay design, fixtures, experiments, reports, and production operation.
Integrations
Adapter kit, AI SDK, Mastra, LangGraph, managed runtimes, Codex, Claude Code, and OpenClaw.
Reference
Exact package, CLI, API, SDK, schema, policy, event, storage, error, and limits reference.
What isplay Covers
Prop
Type
Investigation Loop
Recommended Paths
Use $isplay-analysis to set up isplay for this repo, choose the right adapter, capture my next run, and return an evidence-bounded RCA report.Use this when you want an AI analyst to drive the JSON-first loop and produce a bounded report.
npx isplay start
export ISPLAY_API_URL="<api url shown by isplay start>"
npx isplay projects create --name "Agent Lab"
export ISPLAY_PROJECT_ID="<project id>"
npx isplay discover run <runId>Use this when you want exact command control, CI scripting, or a repeatable manual analysis path.
import { init } from "@isplay/sdk";
const client = init({
projectId: process.env.ISPLAY_PROJECT_ID!,
baseUrl: process.env.ISPLAY_API_URL,
});
await client.withRun({ name: "claim-review" }, async () => {
await client.checkpoint("before-decision", { stage: "ready" });
});Use this when you own the app or framework integration and can instrument the run lifecycle directly.
Where To Go Next
Install
The streamlined install page for skills, adapters, and first capture.
Quickstart
Capture the first run after installation.
First captured run
A guided, concrete path from project creation to captured run evidence.
Object model
The durable objects and how they relate.
Capture design
How to capture enough prompts, tools, state, and checkpoints for useful replay.
Replay and branches
How interventions are applied and how to interpret diffs and comparability.
Fixture handling
How to continue safely when a divergent tool output is missing.
Experiments
Hypothesis batches, trial planning, statistics, effects, and low-N caveats.
Operations
Local stack behavior, Postgres, artifacts, testing, known stubs, and troubleshooting.