Documentation
tracellm replay
Replay a recorded execution timeline.
Overview
The tracellm replay command fetches a saved trace from SQLite and re-executes its timeline step by step in the terminal. Each step is rendered with a tree visualization showing the execution order, status, input, and output.
Tracey Tip
Replay helps you understand agent decisions. Instead of reading logs backwards, you watch the execution tree step by step — seeing exactly which tool was called, what it returned, and how the agent decided to proceed.
Syntax
terminalCopy
bash
tracellm replay [OPTIONS] TRACE_ID
| Argument | Type | Required | Description |
|---|---|---|---|
| TRACE_ID | string | Yes | The trace ID to replay (e.g. tr_2kf9q3m1) |
| Option | Type | Default | Description |
|---|---|---|---|
| --speed | float | 1.0 | Replay speed multiplier (min 0.1) |
| --show-response | flag | false | Print the full saved response after replay |
What Happens Internally
- Fetches the trace from SQLite by ID via
fetch_trace. - Renders metadata — trace ID, status, latency, retry count, and step count in a replay header panel.
- Iterates steps using Rich Live display. For each step it renders an execution tree showing which steps are completed (✓), active (▶), and pending, alongside a detail panel with tool name, duration, status, input (clipped to 200 chars), and output (clipped to 200 chars).
- Throttles timing — waits
duration / 1000 / speedseconds between steps, clamped between 80ms and 550ms. - Prints final report — a summary table plus a response preview (up to 600 chars). If
--show-responseis set, the full response is printed.
Example Output
Terminal outputCopy
text
Replaying execution timeline...
╭── Replay ────────────────────────────────────────────────╮
│ │
│ trace_id tr_2kf9q3m1 │
│ status SUCCESS │
│ latency 3420.00 ms │
│ retries 1 │
│ steps 9 │
│ │
╰──────────────────────────────────────────────────────────╯
╭── Replaying step 3/9 ────────────────────────────────────╮
│ │
│ ╭── Execution Timeline ─────────────────────────────╮ │
│ │ │ │
│ │ ✓ query.embed │ │
│ │ ✓ vector.search │ │
│ │ ▶ context.rerank │ │
│ │ ○ agent.plan │ │
│ │ ○ context.allocate │ │
│ │ ○ tool.chain │ │
│ │ ○ llm.generate │ │
│ │ │ │
│ ╰────────────────────────────────────────────────────╯ │
│ │
│ ╭── Step Detail ────────────────────────────────────╮ │
│ │ │ │
│ │ step 3/9 │ │
│ │ tool context.rerank │ │
│ │ duration 340 ms │ │
│ │ status OK │ │
│ │ input {"query": "Explain transformers",...} │ │
│ │ output {"reranked": true, "matches": 5} │ │
│ │ │ │
│ ╰────────────────────────────────────────────────────╯ │
│ │
╰──────────────────────────────────────────────────────────╯
Replay completeCommon Errors
| Error | Cause | Fix |
|---|---|---|
| Trace not found | Invalid trace ID or trace was deleted | Verify the trace ID with tracellm export or the dashboard |
| No steps to replay | Trace exists but has no recorded steps | Some traces may be empty — check the trace in the dashboard |
| SQLite connection failed | Database file is missing or corrupted | Ensure ~/.tracellm/traces.db exists — it is created automatically on first run |
Troubleshooting
Tip
Use
--speed 2 to replay twice as fast for quick debugging, or --speed 0.5 to watch each step in slow motion.