> ## Documentation Index
> Fetch the complete documentation index at: https://docs.terma.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Replay and Inspect AI Sessions in Terma's Dashboard

> Use Terma's session replay to step through every prompt, tool call, file edit, and model response in a coding session with cost and token context.

Session replay gives you the full timeline of any AI coding session — every prompt sent, every model response received, every file touched, and every tool call made — alongside the token counts and cost that went with it. Whether you're auditing an unexpected bill, debugging a flaky agent run, or understanding why a session ballooned, replay puts the complete record in front of you.

## Navigating to a Session

Open the **Sessions** view in the dashboard. Sessions appear in reverse-chronological order, most recently active first.

Use the filter bar to narrow the list:

* **Person** — filter to a specific engineer or principal
* **Date range** — pick a preset (today, yesterday, last 7 days) or a custom window
* **Source** — limit to a single integration (Claude Code, Codex, Cursor, OpenCode, OpenRouter)
* **Model** — show only sessions that used a given model

Click any row to open the session's replay view.

## Session Timeline

The timeline displays every event in the order it occurred during the session. Each entry shows its timestamp, type, and a one-line summary.

Filter the timeline by event type using the tabs at the top:

| Tab              | What it shows                                                                    |
| ---------------- | -------------------------------------------------------------------------------- |
| **All**          | Every event in the session, unfiltered                                           |
| **Conversation** | Prompts and model responses, in order                                            |
| **Tools**        | Every tool call — file reads, edits, bash commands, and any agent-specific tools |
| **Errors**       | Tool failures, context-window errors, and model refusals                         |

Expand any event to read the full content.

## Token Timeline

The token timeline sits above the event list as a visual chart. It plots token usage across the session so you can spot the moments where context grew large or a prompt spike occurred.

* **Click** any point on the chart to jump to that message in the event list
* **Drag** to select a range and see the aggregate token count for that slice
* The chart annotates the **peak token count** and the model in use at that point

If a session switched models mid-run, each segment is color-coded to its model.

## Key Stats

The stats panel in the top-right corner of the replay view summarizes the whole session at a glance:

| Stat               | Description                         |
| ------------------ | ----------------------------------- |
| **Total messages** | Number of conversational turns      |
| **Tool calls**     | Count of tool invocations           |
| **Prompts**        | Number of user-initiated prompts    |
| **Input tokens**   | Tokens sent to the model            |
| **Output tokens**  | Tokens returned by the model        |
| **Cache read**     | Tokens served from the prompt cache |
| **Cache write**    | Tokens written to the prompt cache  |
| **Cost**           | Total USD cost for the session      |

## Tool Call Detail

Expand any tool call in the **Tools** tab to inspect it fully:

* **Input** — the exact arguments passed to the tool
* **Output** — the full result returned (truncated for large outputs, with a "Show full" option)
* **Duration** — wall-clock time the tool call took
* **Token cost** — input and output tokens attributed to this call

This is useful for spotting expensive file reads, long-running bash commands, or tool calls that returned errors and triggered retry loops.

## Searching Within a Session

Use the search bar at the top of the timeline to find content within the session. Search matches across:

* Message content (prompts and responses)
* Tool call inputs and outputs
* File paths mentioned or edited

Matches are highlighted in the timeline and you can step through them with the arrow keys.

## Full Session View

When a session used context compaction — where the agent summarized earlier context to make room for new turns — the timeline shows only the active window by default. Toggle **Show full session** to expand the view and include compacted turns, so you can trace the complete arc of the work from start to finish.

## From the CLI

The same session data is available in your terminal. Use `terma session events` with the session's routing key:

```bash theme={null}
terma session events 018f3a2c-7d4e-7a1b-9c3d-2e5f6a7b8c9d
```

Add `--tools-only` to focus on tool calls and skip the conversational turns:

```bash theme={null}
terma session events 018f3a2c-7d4e-7a1b-9c3d-2e5f6a7b8c9d --tools-only
```

To find a session's routing key by person, use `terma session list`:

```bash theme={null}
terma session list --user ada --since yesterday
```

<Tip>
  Use `terma session git` with the same routing key to see the git commits and GitHub actions the session performed — useful for linking a session's work to the code it produced.
</Tip>