Session replay
A reconstructable record of an agent or user session — prompts, tool calls, and outcomes — so developers can debug failures and grade task success after the fact.
Learn
When to use it
Use session replay when a simple log of inputs and outputs isn't enough to diagnose issues or understand user interactions. Session replays provide a detailed reconstruction of prompts, tool calls, and outcomes, enabling developers to debug failures and assess task success with insights into things like decision paths and user behavior.
Quick example
In Armature's Product Analytics for AI Agent Sessions, developers can view a session replay to understand why an AI agent failed to complete a task. By examining the sequence of prompts and tool calls, they can identify where the process broke down. Armature includes session replay as a core feature, allowing developers to reconstruct and analyze agent sessions in detail.
prompt → session replay → analyze decisions → debug/fix → stop
Ecosystem
Session replays are part of the debugging and analytics loop, providing context for understanding agent behavior and user interactions.
┌─ analytics ─┐
prompt →│ session replay │→ debug/fix
└─ logs ──────┘
Misconceptions
| Misconception | Rebuttal |
|---|---|
| It captures real-time data | It reconstructs past sessions from logs |
| Only for user interfaces | Also used for AI agent sessions |
| Replays are live streams | They are post-session reconstructions |
Trade-offs
- Detailed insights — requires comprehensive logging and storage
- Debugging aid — potential privacy concerns with sensitive data
- Behavior analysis — can lead to data overload if not managed