AI & agents / current features and roadmap
From replay evidence to agent workflows.
Glowcrumb is AI-native session replay built for understanding and action. Tune capture and analysis, grant an agent access to replay evidence, and explore our MCP and ChatGPT roadmap.
Evidence-linked AI suggestions
Analyze a session on demand to turn possible friction markers into categorized suggestions. Each evidence offset is checked against the markers supplied to the model, so you can return to the relevant moment in the recording.
Connect your own OpenAI, Anthropic, OpenRouter, or public OpenAI-compatible endpoint, choose a model, and set a daily analysis limit. Your provider may charge for requests. Results are cached per session, requests are bounded, and failures count toward the daily limit because a provider may already have billed them.
The analysis uses sanitized interaction signals, not raw DOM recordings. See what AI receives and the other privacy controls before enabling it.
Give an agent scoped access
An agent can read Glowcrumb’s agent setup guide, request a project, and give you an approval link and code. You sign in, review the project and permissions, and approve. Access stays scoped to that project and can be revoked from your workspace.
Installation access includes DNS and script instructions plus domain verification. Separate, optional replay permission lets the agent read recordings and existing insights. Agents cannot change privacy or AI settings, trigger paid analysis, or delete data through this API.
This native registration API is available today. MCP and ChatGPT connectors are the next planned integration layer.
Tune the inputs. Inspect the interpretation.
Configure field and content masking rules, excluded paths, and field-classification instructions for your site. Choose the provider and model used for replay analysis and set its daily limit. Your collection hostname and privacy controls remain part of that foundation.
Useful interpretation connects an observation to a possible explanation and a concrete next step. Glowcrumb links suggestions to supplied replay signals so you can inspect the interpretation in context. A valid timestamp establishes an evidence reference; it does not prove the explanation is correct.
Start with the capture and privacy controls, then check suggestions against recordings before deciding what to change.
From a signal to a next step
- Start with a product question. For example: where might people get stuck during onboarding?
- Find relevant sessions. Use session search and friction filters to narrow your review.
- Analyze a recording. Ask Glowcrumb to generate suggestions from that session’s supplied markers.
- Replay the evidence. Check whether the suggested explanation fits the page and interaction.
- Choose a change to test. An AI suggestion is the beginning of an investigation, not a measured improvement.
MCP and ChatGPT integrations
Our roadmap includes a Model Context Protocol (MCP) integration and a ChatGPT integration, bringing replay investigation closer to your assistants and agents.
The goal is to give agents useful interpretations of website behavior: what happened, the supporting evidence, a problem worth investigating, and a proposed action. MCP is central to that direction, so agents can work with replay context in the tools you choose. These are planned workflows, not capabilities you can connect today.
There is no public Glowcrumb MCP endpoint or ChatGPT connector to install yet. Availability, supported actions, and setup instructions will be documented here when those integrations ship. Using an OpenAI API key for today’s analysis does not connect Glowcrumb to ChatGPT.
Evidence stays central
Repeated clicks can suggest friction. They cannot establish what a visitor felt, whether a button was broken, or whether a proposed fix will help. Review the replay and test changes against your own product goals.
The current product does not autonomously edit or deploy your website. The roadmap is to let agents take bounded, authorized action on findings, with evidence available for review. Agent-driven website changes are not available today.
Start with a verified collection domain, configure privacy and retention, and explore the sample workspace to see how replay context supports a useful investigation.