Replay: interactively step through a recorded execution

Replay is the payoff of recording everything: instead of reproducing a bug, you scrub through what actually happened — every function call, query, log line, and variable, in order, after the fact.

slomo replay SM-8b6f710a      # replay an issue's crash
slomo replay a1b2c3           # replay a session (id prefix ok)
slomo replay                  # default: the latest session

When given an issue id, replay loads the latest incident’s session and positions you in the recorded execution — press t to jump straight to the error. When given a session id, it starts at the beginning of that run.

Commands

Replay is a command loop (type a command, press Enter). Type h anytime for this list:

Command	Action
`n` [N]	next event — or N events forward (`n 10`)
`p` [N]	previous event — or N back
`j N`	jump to event number N
`t`	jump to the next error
`/text`	search forward for text
`?text`	search backward
`i`	inspect the current event’s full payload
`v`	show the variable snapshot linked to the current event
`w`	show a context window of surrounding events
`h`	help
`q`	quit

A typical crash investigation

$ slomo replay SM-8b6f710a
slomo replay — 214 events. Type h for help.

> t                                  # jump to the error
[187] 13:04:52.134  function.exception  TypeError: 'NoneType' object …

> v                                  # locals at the crash
inventory = None
sku       = 'gadget'
db        = <sqlite3.Connection>

> p 3                                # rewind: what led here?
[184] 13:04:52.133  sql.query   SELECT sku, qty FROM inventory WHERE sku = ?

> n                                  # step forward
[185] 13:04:52.133  sql.result  0 rows, 9µs                 ← the smoking gun

> /checkout                          # find where checkout was entered
[180] 13:04:52.132  function.enter  checkout(db=…, sku='gadget')

> i                                  # full payload of the current event
{ "function": "checkout", "module": "app", "file": "app.py", "line": 13,
  "args": ["<sqlite3.Connection>"], "kwargs": {"sku": "gadget"} }

The rhythm: t to the crash, v for the state, then rewind to find where the bad value was produced. Time flows both ways here — that’s the point.

Non-interactive modes

For scripts, CI logs, or piping to other tools:

slomo replay SM-8b6f710a --trace     # print the whole timeline, non-interactively
slomo replay SM-8b6f710a --json      # events as JSON — jq-friendly

slomo replay a1b2c3 --json | jq 'select(.type == "sql.query") | .payload.sql'

Notes

Replay works on the recorded timeline — it re-executes nothing. It is always safe to replay production recordings.
Oversized variable captures are stored as snapshot files and loaded on demand when you press v.
No raw-keypress capture or alternate screen: replay works over SSH, in dumb terminals, and inside tmux without surprises.

Get Started

Core Concepts

Recording

Investigating

Configuration

Replay: interactively step through a recorded execution

Commands

A typical crash investigation

Non-interactive modes

Notes

​Commands

​A typical crash investigation

​Non-interactive modes

​Notes

Commands

A typical crash investigation

Non-interactive modes

Notes