Skip to main content
AI coding agents are good at reading code. They can browse your file tree, trace a call stack, and reason about what your application is supposed to do. What they can’t do — until now — is see what your application actually does when it runs. Subtext closes that gap by connecting your agent to the live signals that have always existed in your app: session recordings, browser devtools, network traffic, and a structured map of your UI.

​
The problem

When a bug gets filed, a design drifts, or a flow breaks in production, a human developer opens a session replay, scrubs to the moment it went wrong, and figures it out. Your agent can’t do that. It has no way to open a recording, inspect a component tree, watch a network request fail, or compare what the UI looked like before and after a change. It works from static context alone — which means it guesses where it should know.

​
The solution

Subtext gives your agent the same visibility a developer has. Install the capture snippet once, connect your agent through the MCP server, and your agent can open real sessions, step through events, take snapshots of your running UI, and check its own work against what actually rendered. It is not a new agent framework or a replacement for your existing tools — it is the layer that makes the tools you already use aware of your running application.

​
Relationship to Fullstory

Subtext is a child brand of Fullstory. The session capture infrastructure — the snippet, the edge recording pipeline, the event stream — is Fullstory’s. Subtext wraps that infrastructure in an agent-native interface: MCP tools, semantic snapshots, and structured workflows your agent can reason about without human translation.

​
Core capabilities

Session Review

Replay real session recordings from your users. Your agent steps through the event timeline, captures UI snapshots at any point, compares states across timestamps, and annotates sessions with threaded comments — no human scrubbing required.

Privacy Governance

Form inputs are masked out of the box. Beyond that, your agent runs a PII scanning loop — flagging elements that carry sensitive data, proposing masking rules, and surfacing them for your approval before they ever leave the browser.

Skills

Slash commands that teach your agent structured workflows. /subtext:review walks your agent through reading the app map, zooming to the moments that matter, and explaining what happened. /subtext:session catalogs every session replay tool and its usage patterns. /subtext:shared establishes the MCP conventions and security rules that all other skills build on.

​
Verify beta

Subtext also includes Verify — a beta set of tools that adds a cloud-hosted browser, artifact storage, proof documents, and the full Sightmap toolchain to your agent’s reach. With Verify, your agent opens a live browser, interacts with your application, inspects the network, and records evidence of its work. See the Verify overview for details.

​
Get started

The fastest way to set up Subtext is the setup wizard. Run it in your project directory and it detects your agent, wires up the capture snippet, configures the MCP server, and installs skills automatically:
Ready to go deeper? Head to the quickstart for a step-by-step walkthrough, or jump straight to the session review reference to see every tool your agent gets.