> ## Documentation Index > Fetch the complete documentation index at: https://docs.pinkfish.ai/llms.txt > Use this file to discover all available pages before exploring further. # Monitor tab > Replay this agent's chats and inspect tool calls from inside the builder The **Monitor** tab inside the Agent Builder is a focused debugger for *this* agent. Where the [org-wide Agent Monitor](/orchestration/agent-monitor) gives you every chat across every agent, the in-builder Monitor only shows chats for the agent you're editing. Monitor tab inside the builder with a real chat selected. Left pane shows the chat list with 'Summarize weekly pipeline changes…' at 4/14/2026 11:34 PM. Right pane shows the full message thread: a user prompt, three 'Tool Call: knowledge_base_search' entries paired with three 'Tool Result: knowledge_base_search' responses, and the agent's reply — each entry timestamped. Use it when you want to answer one of these questions: * *"Why did the agent pick that tool?"* * *"What parameters did it actually pass?"* * *"Did a tool call return an error that the user didn't see in the surfaced reply?"* ## Layout The tab has two panes. ### Left — chat list Every chat this agent has been part of, newest first by default. * **Search here…** — filter the list by chat title. * **Sort order arrow** — flip between newest-first and oldest-first. * Empty state: **"No chats found."** when the search doesn't match, or the agent simply has no chats yet. Click a chat to load it on the right. ### Right — message viewer The full message thread for the selected chat. * **Search messages…** — filter the thread for a keyword. * **Expand all** — expand every tool-call block at once. * Message actions (top-right icons) — copy / download controls for the selected thread. Empty state before you've clicked a chat: > **No Chat Selected** > > *Select a chat from the list to view monitor activity.* If the selected chat has no content yet: > **No Messages Found** > > *This chat doesn't have any messages yet. Start a conversation in the preview panel to see monitor information.* ## Parsed message details The viewer parses each message to surface the agent's internal work, not just the final reply. Expect to see, per message: * The **tool name** the agent called. * The **parameters** it passed. * The **result** (or the error, if the call failed). * The **model's reasoning text** when the model supports it. This is the same content you'd see if you turned on verbose logging, just rendered as a chat replay instead of a log stream. ## When to use this vs the org-wide Monitor | Use the in-builder **Monitor** tab when… | Use the [org-wide **Agent Monitor**](/orchestration/agent-monitor) when… | | ---------------------------------------------------------------------- | ------------------------------------------------------------------------ | | You're iterating on a single agent | You're doing analytics or capacity planning across many agents | | You want to drill from "this run failed" to "this parameter was wrong" | You want to see usage trends or who is chatting with what | | You need to see ephemeral tool errors the user didn't surface | You want to customize columns, date ranges, and export data | Both views are reading the same underlying chat store — switching between them doesn't change any data, just the filter. ## Tips * Pin the Monitor tab before you start iterating. Clicking **+** on the tab strip lets you pin it so you can pivot between Instructions → Tools → Monitor quickly. * While debugging, run the same prompt twice in the Preview pane: once before a change, once after. The two runs sit side-by-side in Monitor, making regressions easy to spot. * If you've published a release, you can see Coworker / Slack / Teams conversations the same way — they're all just chats tied to the agent. ## Related * [Agent Monitor](/orchestration/agent-monitor) — the org-wide version. * [Tools tab](/agents/tools) — where you change what the agent can call once the Monitor tells you it's calling the wrong thing. * [Advanced tab](/agents/advanced) — force structured responses if you're hitting free-text parsing issues.