---
title: ⌨️ CLI manual
order: 1150
---

<!-- DO NOT EDIT! This file was auto-generated by `pixi run man`. -->

## rerun


The Rerun command-line interface:
* Spawn viewers to visualize Rerun recordings and other supported formats.
* Start a gRPC server to share recordings over the network, on native or web.
* Inspect, edit and filter Rerun recordings.


**Usage**: ` rerun [OPTIONS] [URL_OR_PATHS]… [COMMAND]`

**Commands**

* `analytics`: Configure the behavior of our analytics.
* `auth`: Authentication with the redap.
* `download`: Download recordings and save them as .rrd files.
* `man`: Generates the Rerun CLI manual (markdown).
* `mcap`: Manipulate the contents of .mcap files.
* `viewer-mcp`: Run an MCP server that controls a running Rerun Viewer.
* `reset`: Reset the memory of the Rerun Viewer.
* `rrd`: Manipulate the contents of .rrd and .rbl files.
* `server`: In-memory Rerun data server.

**Arguments**

* `<URL_OR_PATHS>`
> Any combination of:
> - A gRPC url to a Rerun server
> - A path to a Rerun .rrd recording
> - A path to a Rerun .rbl blueprint
> - An HTTP(S) URL to an .rrd or .rbl file to load
> - A path to an image or mesh, or any other file that Rerun can load (see https://www.rerun.io/docs/concepts/logging-and-ingestion/importers/overview)
>
> If no arguments are given, a server will be hosted which a Rerun SDK can connect to.

**Options**

* `--bind <BIND>`
> What bind address IP to use.
>
> `::` will listen on all interfaces, IPv6 and IPv4.
>
> [Default: `0.0.0.0`]

* `--memory-limit <MEMORY_LIMIT>`
> An upper limit on how much memory the Rerun Viewer should use.
> When this limit is reached, Rerun will drop the oldest data.
> Example: `16GB` or `50%` (of system total).
> You can also set this in the settings panel.

* `--server-memory-limit <SERVER_MEMORY_LIMIT>`
> An upper limit on how much memory the gRPC server (`--serve-web`) should use.
> The server buffers log messages for the benefit of late-arriving viewers.
> When this limit is reached, Rerun will drop the oldest data.
> Example: `16GB` or `50%` (of system total).
>
> [Default: `1GiB`]

* `--newest-first <NEWEST_FIRST>`
> If true, play back the most recent data first when new clients connect.
>
> [Default: `false`]

* `--cors-allow-origin <CORS_ALLOW_ORIGIN>`
> Additional origin patterns allowed to make CORS requests to the gRPC server.
>
> Use this when hosting a custom viewer on a different domain. Patterns are matched against the full Origin header (e.g. `https://example.com:8080`), using glob-style matching where `*` matches any sequence of characters. Can be specified multiple times.
>
> Examples: `--cors-allow-origin "https://*.example.com"` `--cors-allow-origin "https://example.com:8080"` `--cors-allow-origin "https://example.com:*"`

* `--persist-state <PERSIST_STATE>`
> Whether the Rerun Viewer should persist the state of the viewer to disk.
> When persisted, the state will be stored at the following locations:
> - Linux: `/home/UserName/.local/share/rerun`
> - macOS: `/Users/UserName/Library/Application Support/rerun`
> - Windows: `C:\Users\UserName\AppData\Roaming\rerun`
>
> [Default: `true`]

* `--port <PORT>`
> What port do we listen to for SDKs to connect to over gRPC.
>
> Use `auto` to always start a new viewer with a free port if the default is taken.
>
> [Default: `9876`]

* `--new <NEW>`
> Alias for `--port auto`. Always start a new viewer.
>
> If the port is already in use, a free port will be picked automatically.
>
> [Default: `false`]

* `--profile <PROFILE>`
> Start with the puffin profiler running.
>
> [Default: `false`]

* `--save <SAVE>`
> Stream incoming log events to an .rrd file at the given path.

* `--screenshot-to <SCREENSHOT_TO>`
> Take a screenshot of the app and quit. We use this to generate screenshots of our examples. Useful together with `--window-size`.

* `--serve-web <SERVE_WEB>`
> This will host a web-viewer over HTTP, and a gRPC server, unless one or more URIs are provided that can be viewed directly in the web viewer.
>
> If started, the web server will act like a proxy, listening for incoming connections from logging SDKs, and forwarding it to Rerun viewers.
>
> [Default: `false`]

* `--serve-grpc <SERVE_GRPC>`
> This will host a gRPC server.
>
> The server will act like a proxy, listening for incoming connections from logging SDKs, and forwarding it to Rerun viewers.
>
> [Default: `false`]

* `--connect <CONNECT>`
> Do not attempt to start a new server, instead try to connect to an existing one.
>
> Optionally accepts a URL to a gRPC server.
>
> The scheme must be one of `rerun://`, `rerun+http://`, or `rerun+https://`, and the pathname must be `/proxy`.
>
> The default is `rerun+http://127.0.0.1:9876/proxy`.

* `--expect-data-soon <EXPECT_DATA_SOON>`
> This is a hint that we expect a recording to stream in very soon.
>
> This is set by the `spawn()` method in our logging SDK.
>
> The viewer will respond by fading in the welcome screen, instead of showing it directly. This ensures that it won't blink for a few frames before switching to the recording.
>
> [Default: `false`]

* `-j, --threads <THREADS>`
> The number of compute threads to use.
>
> If zero, the same number of threads as the number of cores will be used. If negative, will use that much fewer threads than cores.
>
> Rerun will still use some additional threads for I/O.
>
> [Default: `-2`]

* `--version <VERSION>`
> Print version and quit.
>
> [Default: `false`]

* `--web-viewer <WEB_VIEWER>`
> Start the viewer in the browser (instead of locally).
>
> Requires Rerun to have been compiled with the `web_viewer` feature.
>
> This implies `--serve-web`.
>
> [Default: `false`]

* `--web-viewer-port <WEB_VIEWER_PORT>`
> What port do we listen to for hosting the web viewer over HTTP. A port of 0 will pick a random port.
>
> [Default: `9090`]

* `--hide-welcome-screen <HIDE_WELCOME_SCREEN>`
> Hide the normal Rerun welcome screen.
>
> [Default: `false`]

* `--detach-process <DETACH_PROCESS>`
> Detach the native Rerun Viewer process from the invoking process.
>
> Ignored for any command that doesn't spawn a viewer.
>
> [Default: `false`]

* `--headless <HEADLESS>`
> Run the viewer in headless mode (no OS window).
>
> The viewer is driven by an offscreen `egui_kittest` harness, while the gRPC server keeps running so SDK clients can still log data and request screenshots via `save_screenshot`.
>
> [Default: `false`]

* `--window-size <WINDOW_SIZE>`
> Set the screen resolution (in logical points), e.g. "1920x1080". Useful together with `--screenshot-to`.

* `--renderer <RENDERER>`
> Override the default graphics backend and for a specific one instead.
>
> When using `--web-viewer` this should be one of: `webgpu`, `webgl`.
>
> When starting a native viewer instead this should be one of:
>
> * `vulkan` (Linux & Windows only)
>
> * `gl` (Linux & Windows only)
>
> * `metal` (macOS only)

* `--video-decoder <VIDEO_DECODER>`
> Overwrites hardware acceleration option for video decoding.
>
> By default uses the last provided setting, which is `auto` if never configured.
>
> Depending on the decoder backend, these settings are merely hints and may be ignored.
> However, they can be useful in some situations to work around issues.
>
> Possible values:
>
> * `auto`
>   May use hardware acceleration if available and compatible with the codec.
>
> * `prefer_software`
>   Should use a software decoder even if hardware acceleration is available.
>   If no software decoder is present, this may cause decoding to fail.
>
> * `prefer_hardware`
>   Should use a hardware decoder.
>   If no hardware decoder is present, this may cause decoding to fail.

* `--test-receive <TEST_RECEIVE>`
> Ingest data and then quit once the goodbye message has been received.
>
> Used for testing together with `RERUN_PANIC_ON_WARN=1`.
>
> Fails if no messages are received, or if no messages are received within a dozen or so seconds.
>
> [Default: `false`]

## rerun analytics

Configure the behavior of our analytics.

**Usage**: `rerun analytics <COMMAND>`

**Commands**

* `details`: Prints extra information about analytics.
* `clear`: Deletes everything related to analytics.
* `email`: Associate an email address with the current user.
* `enable`: Enable analytics.
* `disable`: Disable analytics.
* `config`: Prints the current configuration.

## rerun analytics email

Associate an email address with the current user.

**Usage**: `rerun analytics email <EMAIL>`

**Arguments**

* `<EMAIL>`

## rerun auth

Authentication with the redap.

**Usage**: `rerun auth <COMMAND>`

**Commands**

* `login`: Log into Rerun.
* `logout`: Log out of Rerun.
* `token`: Retrieve the stored access token.
* `generate-token`: Generate a fresh access token.

## rerun auth login

Log into Rerun.

This command opens a page in your default browser, allowing you to log in to Rerun Hub.

Once you've logged in, your credentials are stored on your machine.

To sign up, contact us through the form linked at <https://rerun.io/#open-source-vs-commercial>.

**Usage**: `rerun auth login [OPTIONS]`

**Options**

* `--no-open-browser <NO_OPEN_BROWSER>`
> Post a link instead of directly opening in the browser.
>
> [Default: `false`]

* `--force <FORCE>`
> Trigger the full login flow even if valid credentials already exist.
>
> [Default: `false`]

## rerun auth logout

Log out of Rerun.

This command clears the credentials stored on your machine and ends your session.

**Usage**: `rerun auth logout [OPTIONS]`

**Options**

* `--no-open-browser <NO_OPEN_BROWSER>`
> Post a link instead of directly opening in the browser.
>
> [Default: `false`]

## rerun auth generate-token

Generate a fresh access token.

You can use this token to authorize requests to Rerun Hub.

It's closer to an API key than an access token, as it can be revoked before it expires.

**Usage**: `rerun auth generate-token [OPTIONS] --server <SERVER> --expiration <EXPIRATION>`

**Options**

* `--server <SERVER>`
> Origin of the server to request the token from.

* `--expiration <EXPIRATION>`
> Duration of the token, either in: - "human time", e.g. `1 day`, or - ISO 8601 duration format, e.g. `P1D`.

* `--permission <PERMISSION>`
> Which permission the token should have.
>
> [`read`, `read-write`]
>
> [Default: `read`]

## rerun download

Download recordings and save them as .rrd files.

Supports downloading from Rerun Hub as well as any other supported URI.

**Usage**: `rerun download [OPTIONS] <URLS>…`

**Arguments**

* `<URLS>`
> One or more URIs to download.

**Options**

* `-o, --output-dir <OUTPUT_DIR>`
> Override the output directory for the downloaded `.rrd` files.
>
> Defaults to the current working directory.

## rerun mcap

Manipulate the contents of .mcap files.

**Usage**: `rerun mcap <COMMAND>`

**Commands**

* `convert`: Convert an .mcap file to an .rrd.
* `info`: Print recording, channel, and compression information for an .mcap file.
* `check`: Check an .mcap file for structural and timeline issues.

## rerun mcap convert

Convert an .mcap file to an .rrd.

**Usage**: `rerun mcap convert [OPTIONS] <PATH_TO_INPUT_MCAP>`

**Arguments**

* `<PATH_TO_INPUT_MCAP>`
> Paths to read from. Reads from standard input if none are specified.

**Options**

* `-o, --output <dst.rrd>`
> Path to write to. Writes to standard output if unspecified.

* `--application-id <APPLICATION_ID>`
> If set, specifies the application id of the output.

* `-d, --decoder <SELECTED_DECODERS>`
> Specifies which decoders to apply during conversion.

* `--disable-raw-fallback <DISABLE_RAW_FALLBACK>`
> Disable using the raw decoder as a fallback for unsupported channels. By default, channels that cannot be handled by semantic decoders (protobuf, ROS2) will be processed by the raw decoder.
>
> [Default: `false`]

* `--recording-id <RECORDING_ID>`
> If set, specifies the recording id of the output.
>
> When this flag is set and multiple input .rdd files are specified, blueprint activation commands will be dropped from the resulting output.

* `--timestamp-offset-ns <TIMESTAMP_OFFSET_NS>`
> If set, an offset in nanoseconds to add to all timestamp timelines.
>
> This can be used to shift all timestamps of the MCAP file if they are not yet relative to the UNIX epoch.
>
> Duration and sequence timelines are not affected by this offset.

* `--timeline-type <TIMELINE_TYPE>`
> The timeline type to use for timestamp timelines.
>
> "timestamp" (default) creates `TimestampNs` timelines (nanoseconds since Unix epoch). "duration" creates `DurationNs` timelines (nanosecond durations).
>
> [Default: `timestamp`]

* `-y, --include-topic-regex <INCLUDE_TOPIC_REGEX>`
> Include only topics matching this regex (RE2 syntax). Repeatable.
>
> If omitted, all topics are included. Patterns are not implicitly anchored; use `^` / `$` if you need anchoring.
>
> Example: `-y "^/tf.*" -n ".*depth.*" -y "^/camera/(compressed|camera_info)$"`

* `-n, --exclude-topic-regex <EXCLUDE_TOPIC_REGEX>`
> Exclude topics matching this regex (RE2 syntax). Repeatable.
>
> Applied after includes: a topic is kept only if it matches an include (or no includes are set) AND matches no exclude.

* `--start-time <TIME>`
> Inclusive lower bound on the raw MCAP `log_time`.
>
> Accepts Unix timestamps with a unit suffix (`ns`, `ms`, `s`, …), or an RFC 3339 timestamp. Bare integers are interpreted as nanoseconds.
>
> If set, only data within this time range gets converted.

* `--end-time <TIME>`
> Exclusive upper bound on the raw MCAP `log_time`.
>
> Accepts Unix timestamps with a unit suffix (`ns`, `ms`, `s`, …), or an RFC 3339 timestamp. Bare integers are interpreted as nanoseconds.
>
> If set, only data within this time range gets converted.

* `--recover <RECOVER>`
> Recover a missing or invalid MCAP summary in memory.
>
> This allows conversion of MCAP files that lack a footer (e.g. corrupted recordings).
>
> [Default: `false`]

## rerun mcap info

Print recording, channel, and compression information for an .mcap file.

**Usage**: `rerun mcap info [OPTIONS] <PATH>`

**Arguments**

* `<PATH>`
> Path to the .mcap file to inspect.

**Options**

* `--recover <RECOVER>`
> Reconstruct a missing or invalid summary from the readable portion of the file.
>
> [Default: `false`]

## rerun mcap check

Check an .mcap file for structural and timeline issues.

Reports timelines that disagree on row ordering, whole-topic ordering conflicts, and chunks that arrive out of order on a timeline.

**Usage**: `rerun mcap check [OPTIONS] <PATH>`

**Arguments**

* `<PATH>`
> Path to the .mcap file to check.

**Options**

* `--full <FULL>`
> Check timelines produced by the full `re_mcap` decoder pipeline.
>
> This decodes message payloads to include derived timelines, such as `ros2_timestamp` for ROS 2 data, and increases processing time.
>
> [Default: `false`]

* `--recover <RECOVER>`
> Reconstruct a missing or invalid summary from the readable portion of the file.
>
> [Default: `false`]

## rerun rrd

Manipulate the contents of .rrd and .rbl files.

**Usage**: `rerun rrd <COMMAND>`

**Commands**

* `compare`: Compares the data between 2 .rrd files, returning a successful shell exit code if they match.
* `filter`: Filters out data from .rrd/.rbl files/streams, and writes the result to standard output.
* `merge`: Merges the contents of multiple .rrd/.rbl files/streams, and writes the result to standard output.
* `migrate`: Migrate one or more .rrd files to the newest Rerun version.
* `optimize`: Optimizes the contents of one or more .rrd/.rbl files/streams by compacting chunks, and writes the result to standard output.
* `print`: Print the contents of one or more .rrd/.rbl files/streams.
* `route`: Manipulates the metadata of log message streams without decoding the payloads.
* `split`: Optimally splits a recording on a specified timeline.
* `stats`: Compute important statistics for one or more .rrd/.rbl files/streams.
* `verify`: Verify the that the .rrd file can be loaded and correctly interpreted.

## rerun rrd compare

Compares the data between 2 .rrd files, returning a successful shell exit code if they match.

This ignores the `log_time` timeline.

**Usage**: `rerun rrd compare [OPTIONS] <PATH_TO_RRD1> <PATH_TO_RRD2>`

**Arguments**

* `<PATH_TO_RRD1>`

* `<PATH_TO_RRD2>`

**Options**

* `--unordered <UNORDERED>`
> If specified, the comparison will focus purely on semantics, ignoring order.
>
> The Rerun data model is itself unordered, and because many of the internal pipelines are asynchronous by nature, it is very easy to end up with semantically identical, but differently ordered data. In most cases, the distinction is irrelevant, and you'd rather the comparison succeeds.
>
> [Default: `false`]

* `--full-dump <FULL_DUMP>`
> If specified, dumps both .rrd files as tables.
>
> [Default: `false`]

* `--ignore-chunks-without-components <IGNORE_CHUNKS_WITHOUT_COMPONENTS>`
> If specified, the comparison will ignore chunks without components.
>
> [Default: `false`]

* `--ignore-timeline <TIMELINE>`
> Timelines to ignore entirely during comparison (their presence, absence, and values).
>
> Useful when comparing recordings produced with different default-timeline settings, e.g. `--ignore-timeline log_tick` (which is opt-in). Can be specified multiple times.

## rerun rrd filter

Filters out data from .rrd/.rbl files/streams, and writes the result to standard output.

Reads from standard input if no paths are specified.

This will not affect the chunking of the data in any way.

Example: `rerun rrd filter --drop-timeline log_tick /my/recordings/*.rrd > output.rrd`

**Usage**: `rerun rrd filter [OPTIONS] [PATH_TO_INPUT_RRDS]…`

**Arguments**

* `<PATH_TO_INPUT_RRDS>`
> Paths to read from. Reads from standard input if none are specified.

**Options**

* `-o, --output <dst.(rrd|rbl)>`
> Path to write to. Writes to standard output if unspecified.

* `--drop-timeline <DROPPED_TIMELINES>`
> Names of the timelines to be filtered out.

* `--drop-entity <DROPPED_ENTITY_PATHS>`
> Paths of the entities to be filtered out.

* `--continue-on-error <CONTINUE_ON_ERROR>`
> If set, will try to proceed even in the face of IO and/or decoding errors in the input data.
>
> [Default: `false`]

## rerun rrd merge

Merges the contents of multiple .rrd/.rbl files/streams, and writes the result to standard output.

Reads from standard input if no paths are specified.

⚠️ This will automatically migrate the data to the latest version of the RRD protocol, if needed. ⚠️

Example: `rerun rrd merge /my/recordings/*.rrd > output.rrd`

**Usage**: `rerun rrd merge [OPTIONS] [PATH_TO_INPUT_RRDS]…`

**Arguments**

* `<PATH_TO_INPUT_RRDS>`
> Paths to read from. Reads from standard input if none are specified.

**Options**

* `-o, --output <dst.(rrd|rbl)>`
> Path to write to. Writes to standard output if unspecified.

* `--continue-on-error <CONTINUE_ON_ERROR>`
> If set, will try to proceed even in the face of IO and/or decoding errors in the input data.
>
> [Default: `false`]

## rerun rrd migrate

Migrate one or more .rrd files to the newest Rerun version.

Example: `rerun rrd migrate foo.rrd` Results in a `foo.backup.rrd` (copy of the old file) and a new `foo.rrd` (migrated).

**Usage**: `rerun rrd migrate [PATH_TO_INPUT_RRDS]…`

**Arguments**

* `<PATH_TO_INPUT_RRDS>`
> Paths to rrd files to migrate.

## rerun rrd optimize

Optimizes the contents of one or more .rrd/.rbl files/streams by compacting chunks, and writes the result to standard output.

Reads from standard input if no paths are specified.

If any input is a directory, the command switches to **directory mirror mode**: every `.rrd`/`.rbl` file under the input is optimized independently, and written to the output path while preserving the input folder structure. In this mode the output (`-o`) must be set and is treated as a directory root.

Uses the usual environment variables to control the compaction thresholds: `RERUN_CHUNK_MAX_ROWS`, `RERUN_CHUNK_MAX_ROWS_IF_UNSORTED`, `RERUN_CHUNK_MAX_BYTES`.

Unless explicit flags are passed, in which case they will override environment values.

Video stream chunks are also rebatched on GoP (keyframe) boundaries so that each chunk holds one or more complete GoPs. Pass `--no-rebatch-videos` to disable that.

⚠️ This will automatically migrate the data to the latest version of the RRD protocol, if needed. ⚠️

Examples:

* Optimize a single recording into one optimized file (`-o`): `rerun rrd optimize my.rrd -o my-compacted.rrd`

* Merge many recordings into one optimized file: `rerun rrd optimize --max-size 2MiB /my/recordings/*.rrd -o output.rrd`

* Pipe through stdin/stdout, overriding both row and size thresholds: `cat my.rrd | rerun rrd optimize --max-rows 4096 --max-size 2MiB > output.rrd`

* Directory mirror mode — optimize every `.rrd`/`.rbl` under a tree, preserving structure: `rerun rrd optimize --max-size 2MiB /my/recordings -o /my/recordings-compacted`

**Usage**: `rerun rrd optimize [OPTIONS] [PATH_TO_INPUT_RRDS]…`

**Arguments**

* `<PATH_TO_INPUT_RRDS>`
> Paths to read from. Reads from standard input if none are specified.

**Options**

* `-o, --output <dst.(rrd|rbl)>`
> Path to write the optimized recording to.
>
> In single-file mode (the default), this is the output file path. If unspecified, the recording is written to standard output.
>
> In directory mirror mode (when any input is a directory), this must be set and is treated as the output directory root: the input folder structure is mirrored underneath it, with each `.rrd`/`.rbl` file optimized independently.

* `--profile <PROFILE>`
> Optimization profile to start from.
>
> Per-knob flags and `RERUN_CHUNK_MAX_*` env vars override the profile's values. `RERUN_STORE_ENABLE_CHANGELOG` is ignored by this command — `rerun rrd optimize` is always headless.
>
> [Default: `object-store`]

* `--max-size <MAX_SIZE>`
> Threshold after which a Chunk cannot be compacted any further.
>
> Accepts a size string with a unit suffix, e.g. `2MiB`, `512KiB`, `1GB`, `1024B`. Both binary (`KiB`/`MiB`/`GiB`/`TiB`) and decimal (`kB`/`MB`/`GB`/`TB`) units are accepted.
>
> Overrides the profile's value and `RERUN_CHUNK_MAX_BYTES` if set.

* `--max-rows <MAX_ROWS>`
> What is the threshold, in rows, after which a Chunk cannot be compacted any further?
>
> Overrides the profile's value and `RERUN_CHUNK_MAX_ROWS` if set.

* `--max-rows-if-unsorted <MAX_ROWS_IF_UNSORTED>`
> What is the threshold, in rows, after which a Chunk cannot be compacted any further?
>
> This specifically applies to _non_ time-sorted chunks.
>
> Overrides the profile's value and `RERUN_CHUNK_MAX_ROWS_IF_UNSORTED` if set.

* `--num-pass <NUM_EXTRA_PASSES>`
> Configures the number of extra compaction passes to run on the data. Overrides the profile's value. Default per profile: 50.
>
> Compaction in Rerun is an iterative, convergent process: every single pass will improve the quality of the compaction (with diminishing returns), until it eventually converges into a stable state. The more passes, the better the compaction quality.
>
> Under the hood, you can think of it as a kind of clustering algorithm: every incoming chunk finds the most appropriate chunk to merge into, thereby creating a new cluster, which is itself just a bigger chunk. On the next pass, these new clustered chunks will themselves look for other clusters to merge into, yielding even bigger clusters, which again are also just chunks. And so on and so forth.
>
> If/When the data reaches a stable optimum, the computation will stop immediately, regardless of how many passes are left.

* `--continue-on-error <CONTINUE_ON_ERROR>`
> If set, will try to proceed even in the face of IO and/or decoding errors in the input data.
>
> [Default: `false`]

* `--no-rebatch-videos <NO_REBATCH_VIDEOS>`
> Disable rebatching of video stream chunks to GoP (Group of Pictures) boundaries.
>
> By default, after compaction, video stream chunks are rebatched on GoP boundaries so that each chunk contains one or more complete GoPs. This flag disables that behavior.
>
> Note: GoP rebatching never splits a GoP across chunks, so streams with long keyframe intervals (e.g. 10+ seconds between I-frames) can produce chunks much larger than `--max-size`.
>
> [Default: `false`]

* `--fix-keyframe <FIX_KEYFRAME>`
> Drop any user-supplied `VideoStream:is_keyframe` labels and re-derive them from the encoded samples.
>
> By default, `rrd optimize` validates user-supplied keyframe labels against the encoded samples and errors out if they disagree. Pass this flag to ignore the existing labels and unconditionally re-derive them.
>
> [Default: `false`]

* `--split-size-ratio <SPLIT_SIZE_RATIO>`
> If set, split chunks so no two archetype groups sharing a chunk differ in byte size by more than this factor. Values should be `>= 1`; at `1.0`, every archetype is forced into its own chunk.
>
> This keeps "thick" columns (images, videos, blobs) out of the same chunk as "thin" columns (scalars, transforms, text), so the viewer can fetch just the thin data without dragging along the thick payload. Components belonging to the same archetype are always kept together.
>
> A good starting value is 10.0. If unset, the profile's value is used.

## rerun rrd print

Print the contents of one or more .rrd/.rbl files/streams.

Reads from standard input if no paths are specified.

Example: `rerun rrd print /my/recordings/*.rrd`

**Usage**: `rerun rrd print [OPTIONS] [PATH_TO_INPUT_RRDS]…`

**Arguments**

* `<PATH_TO_INPUT_RRDS>`
> Paths to read from. Reads from standard input if none are specified.

**Options**

* `-v, --verbose <VERBOSE>`
> If set, print out table contents.
>
> This can be specified more than once to toggle more and more verbose levels (e.g. -vvv):
>
> * default: summary with short names.
>
> * `-v`: summary with fully-qualified names.
>
> * `-vv`: show all chunk metadata headers, keep the data hidden.
>
> * `-vvv`: show all chunk metadata headers as well as the data itself.
>
> [Default: `0`]

* `--continue-on-error <CONTINUE_ON_ERROR>`
> If set, will try to proceed even in the face of IO and/or decoding errors in the input data.

* `--migrate <MIGRATE>`
> Migrate chunks to latest version before printing?

* `--full-metadata <FULL_METADATA>`
> If true, includes `rerun.` prefixes on keys.

* `--entity <ENTITY>`
> Show only chunks belonging to this entity.

* `--footers <FOOTERS>`
> If true, displays all the parsed footers at the end.

* `--footers-lod <FOOTERS_LOD>`
> The level of detail to use when printing footers. Higher is more detailed.
>
> * `0`: only chunk metadata columns
>
> * `1`: `0` + global timeline columns
>
> * `2`: `1` + everything else
>
> [Default: `0`]

* `--transposed <TRANSPOSED>`
> Transpose record batches before printing them?

## rerun rrd route

Manipulates the metadata of log message streams without decoding the payloads.

This can be used to combine multiple .rrd files into a single recording. Example: `rerun rrd route --recording-id my_recording /my/recordings/*.rrd > output.rrd`

Note: Because the payload of the messages is never decoded, no migration or verification will performed.

**Usage**: `rerun rrd route [OPTIONS] [PATH_TO_INPUT_RRDS]…`

**Arguments**

* `<PATH_TO_INPUT_RRDS>`
> Paths to read from. Reads from standard input if none are specified.

**Options**

* `-o, --output <dst.rrd>`
> Path to write to. Writes to standard output if unspecified.

* `--continue-on-error <CONTINUE_ON_ERROR>`
> If set, will try to proceed even in the face of IO and/or decoding errors in the input data.
>
> [Default: `false`]

* `--application-id <APPLICATION_ID>`
> If set, specifies the application id of the output.

* `--recording-id <RECORDING_ID>`
> If set, specifies the recording id of the output.
>
> When this flag is set and multiple input .rdd files are specified, blueprint activation commands will be dropped from the resulting output.

## rerun rrd split

Optimally splits a recording on a specified timeline.

The sum of the generated splits will always exactly match the original recording.

Example: `rerun rrd split --output-dir ./splits --timeline log_tick --time 33 --time 66 ./my_video.rrd`

**Usage**: `rerun rrd split [OPTIONS] --output-dir <output directory> --timeline <TIMELINE> <PATH_TO_INPUT_RRD>`

**Arguments**

* `<PATH_TO_INPUT_RRD>`
> Path to read from.

**Options**

* `-o, --output-dir <output directory>`
> Path to the output directory. All generated RRD files will end up there.

* `--timeline <TIMELINE>`
> The timeline used to compute the splits.
>
> The other timelines will be kept in the output, which might or might not make sense depending on the density of the dataset. Use `--drop-unused-timelines` to discard them.

* `-t, --time <TIMES>`
> The timestamps at which to perform the splits. Incompatible with `--num-parts`/`-n`.
>
> There are always `number_of_times + 1` resulting splits.
>
> For example, given `-t 10 -t 20 -t 30`, this command will output 4 splits: [-inf:10), [10:20), [20:30), [30:+inf).

* `-n, --num-parts <NUM_PARTS>`
> The number of parts to split the recording into. Incompatible with `--time`/`-t`.
>
> There will be exactly that number of resulting splits. Each split will cover an equal time span in the timeline.

* `--recording-id <recording ID prefix>`
> The recording ID prefix to be used for the output recordings.
>
> If left unspecified, the ID of the original recording, suffixed with a `-`, will be used as a prefix.
>
> Each split will use `<recording_id_prefix><i>` as their respective recording ID, where `i` is the index of the split.

* `--drop-unused-timelines <DISCARD_UNUSED_TIMELINES>`
> If true, timelines other than the one specified with `--timeline` will be discarded.
>
> [Default: `false`]

## rerun rrd stats

Compute important statistics for one or more .rrd/.rbl files/streams.

Reads from standard input if no paths are specified.

Example: `rerun rrd stats /my/recordings/*.rrd`

**Usage**: `rerun rrd stats [OPTIONS] [PATH_TO_INPUT_RRDS]…`

**Arguments**

* `<PATH_TO_INPUT_RRDS>`
> Paths to read from. Reads from standard input if none are specified.

**Options**

* `--no-decode <NO_DECODE>`
> If set, the data will never be decoded.
>
> Statistics will be computed at the transport-level instead, which is more limited in terms of what can be computed, but also orders of magnitude faster.
>
> [Default: `false`]

* `--continue-on-error <CONTINUE_ON_ERROR>`
> If set, will try to proceed even in the face of IO and/or decoding errors in the input data.
>
> [Default: `true`]

## rerun rrd verify

Verify the that the .rrd file can be loaded and correctly interpreted.

Can be used to ensure that the current Rerun version can load the data.

**Usage**: `rerun rrd verify [OPTIONS] [PATH_TO_INPUT_RRDS]…`

**Arguments**

* `<PATH_TO_INPUT_RRDS>`
> Paths to read from. Reads from standard input if none are specified.

**Options**

* `--check-footers <CHECK_FOOTERS>`
> If true, ensures that RRD footers are present and well formed.
>
> [Default: `true`]

## rerun server

In-memory Rerun data server.

**Usage**: `rerun server [OPTIONS]`

**Options**

* `--host <HOST>`
> IP address to listen on.
>
> [Default: `0.0.0.0`]

* `-p, --port <PORT>`
> Port to bind to.
>
> [Default: `51234`]

* `-d, --dataset <[NAME=]DIR_PATH>`
> Load a directory of RRD as dataset (can be specified multiple times). You can specify only a path or provide a name such as `-d my_dataset=./path/to/files`.

* `-t, --table <[NAME=]TABLE_PATH>`
> Load a lance file as a table (can be specified multiple times). You can specify only a path or provide a name such as `-t my_table=./path/to/table`.

* `--latency-ms <LATENCY_MS>`
> Artificial latency to add to each request (in milliseconds).
>
> [Default: `0`]

* `--bandwidth-limit <BANDWIDTH_LIMIT>`
> Artificial bandwidth limit for responses (e.g. '10MB' for 10 megabytes per second).

* `--cors-allow-origin <CORS_ALLOW_ORIGIN>`
> Additional origin patterns allowed to make cross-origin requests to the server (can be specified multiple times).
>
> By default, only `localhost`, `127.0.0.1`, and `rerun.io` are allowed. Patterns are matched against the full `Origin` header value, using glob-style matching where `*` matches any sequence of characters.

* `-V, --version `
> Print version.