Is there any documentation on webviewer callback? I see that they are available here, but just see a single example. Is there a list of callbacks with function call signature, etc anywhere? Is the callback structure stable or should we treat it as a private API that changes without warning? https://rerun.io/docs/howto/integrations/embed-web#callbacks

Callbacks 
The Viewer API also allows registering callbacks for certain events.

For example, here is how you would react to entities being selected in the Viewer:


viewer.on("selection_change", (event) => {
  for (const item of event.items) {
    if (item.type === "entity") {
      console.log(item.entity_path);
    }
  }
});

Previous to rerun version 0.23, I did run the rerun webviewer (served over the cli) and forwarded the ws port with nginx. That way I could access the webviewer with an open websocket connection e.g. via: http://172.31.11.40/mycluster/rerun/?url=ws%3A%2F%2F172.31.11.40%2Fmycluster%2Fws

location $BASE_PATH/rerun/ {
        proxy_pass http://localhost:9090/;
        proxy_http_version 1.1;
        proxy_set_header Upgrade $http_upgrade;
        proxy_set_header Connection "Upgrade";
        proxy_set_header Host $host;
        proxy_set_header X-Real-IP $remote_addr;
        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
        proxy_set_header X-Forwarded-Proto $scheme;
    }

    location = $BASE_PATH/ws {
        return 301 $scheme://$http_host$BASE_PATH/ws/;
    }

    location $BASE_PATH/ws/ {
        proxy_pass http://localhost:9877/;
        proxy_http_version 1.1;
        proxy_set_header Upgrade $http_upgrade;
        proxy_set_header Connection "Upgrade";
        proxy_set_header Host $host;
        proxy_set_header X-Real-IP $remote_addr;
        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
        proxy_set_header X-Forwarded-Proto $scheme;
    }

As of 0.23, the connection switched to grpc and I can't make it work the same way. How can I get rerun running inside the pod? My goal is to connect a vscode webviewer which can execute python code which logs to a rerun web viewer instance. Using rrd files would work but does not push data (I don't want to refresh the view).

Hi rerun team, I have a question regarding the use of the columnar api when the data have some gaps. So, in my case, I have several archetypes that I need to log (boxes, 3dpoints line series...) but all of them might have gaps in the recording. Now, with the row api, I'd just call clear on that entity when it is not available. Unfortunately, due to performance issues, I need to use the columnar api. How can I clear the content in this case? I came up with 2 solutions: - use the columnar api and log NaN values on the components of the archetype, which basically makes it disappear until a valid value is logged again. - log all valid entries with the columnar api, and use the row api to clear content on that entity using clear/cleared when needed both are not great, the first one seems like a hack, and the second one is a bit cumbersome to implement if there are many gaps. Can you think of some other way to achieve this? what would you suggest? Thanks!

Here is my current scenario: I have a CLI with an option to enable and disable Rerun logging. I also want to disable Rerun logging if I am unable to connect to a valid viewer. In order to do this I am relying on calls to

RecordingStream

to "early out" when disabled. However, I'm finding that

connect_tcp

doesn't seem to require a connection to a valid viewer for

connect_tcp(url).is_ok()

to return

ErrorCode::Ok

. For instance, if I don't open a Rerun viewer and run my application, I get the following:

CLI OPTION SET: Rerun viewer IP:PORT = 127.0.0.1:9876
[2025-06-18T18:15:12Z WARN  re_sdk_comms::buffered_client] Failed to send message after 3 attempts: Failed to connect to Rerun server at 127.0.0.1:9876: Connection refused (os error 111)
[2025-06-18T18:15:15Z WARN  re_sdk_comms::buffered_client] Dropping messages because tcp client has timed out.
[2025-06-18T18:15:15Z WARN  re_sdk_comms::buffered_client] Dropping messages because tcp client has timed out.
[2025-06-18T18:15:15Z WARN  re_sdk_comms::tcp_client] Tried to flush while TCP stream was still Pending. Data was possibly dropped.
Connected to 127.0.0.1:9876
Rerun logging enabled.

I have a largish-number og rrd files and would like to explore some of them based on some data I have available in a dataframe. As such, it would be convienent to have a function in the SDK or an option from the cli to not launch a new instance of the viewer, but ask the existing view to open a specific file

Is there a way to easily use the browser to view rerun streams? To save wifi bandwidth I want to run the viewer locally on my robot and ideally use a web page to see the data rendered on the robot when I need to. Otherwise, is there an alternative way? basically I would like to be able to opt in when to stream the data out of the robot, but keep recording it in the background. Thank you

Hello, I am observing a lot of latency often sporadically that gradually builds up. I already use a memory limit when spawning the viewer. But this does not help with latency building since this is during the logging time. I also noticed

--drop-at-latency

flag: https://rerun.io/docs/howto/visualization/limit-ram#dropatlatency in the documentation, but what is the way to incorporate this using C++ similar to memory limit spawn options?

When I include rr.save('adslfkj.rrd') I no longer see any data streamed live to the viewer. I can later open up the .rrd file but that's a bit annoying. Can I view the live data and write to file at the same time?

We are using the C++ sdk with c++20 std. Used valgrind massif to log the heap allocations. With rerun logging on I am seeing unbounded growth and with it off I am seeing a flat profile which is expected. We have rerun actively logging to the viewer over the network and are seeing data showing up. We are filling up 8 gigs of ram in less than 30 seconds when we are not profiling. I am using version 0.22.1 include(FetchContent) FetchContent_Declare( rerun_sdk URL https://github.com/rerun-io/rerun/releases/download/0.22.1/rerun_cpp_sdk.zip ) FetchContent_MakeAvailable(rerun_sdk) I am not sure if this issue is relevant but it has no reply https://github.com/mozilla/uniffi-rs/issues/2470 Any thoughts or suggestions are appreciated! https://cdn.discordapp.com/attachments/1380251130340315146/1380251131027914793/L824KW8tjiEcBAAAAAAAAAAAAAAAAAAAAAAAAAPgS2BYqflFizSdhKtDc9ZjHcd2T7LvrPRKPAgAAAAAAAAAAAAAAAAAAAAAAAACArU46UfM8Oc2zlcecJj4t998WkW4757ZpscSjAAAAAAAAAAAAAAAAAAAAAAAAAAAAp3TWsPYsx28KSIlHAQAAAAAAAAAAAAAAAAAAAAAAAAAAzqlxfGpmj81mc2jkQWyfgAAAABJRU5ErkJggg.png?ex=6853ad49&is=68525bc9&hm=cd889c185502cfa362d0d20906aced17b9269394de874d64d60bc6180a92a1b9& https://cdn.discordapp.com/attachments/1380251130340315146/1380251131816579142/L824KW8tjiEcBAAAAAAAAAAAAAAAAAAAAAAAAAPgS2BYqflFizSdhKtDc9ZjHcd2T7LvrPRKPAgAAAAAAAAAAAAAAAAAAAAAAAACArU46UfM8Oc2zlcecJj4t998WkW4757ZpscSjAAAAAAAAAAAAAAAAAAAAAAAAAAAAp3TWsPYsx28KSIlHAQAAAAAAAAAAAAAAAAAAAAAAAAAAzqlxfGpmj81mc2jkQWyfgAAAABJRU5ErkJggg.png?ex=6853ad49&is=68525bc9&hm=e16ec250f510150646db17784a3c1097574762af6a2e6255aae914fca4b51243& https://cdn.discordapp.com/attachments/1380251130340315146/1380251132651114627/5K2BmM6gAAAABJRU5ErkJggg.png?ex=6853ad49&is=68525bc9&hm=b83de04f18164540e1672bb3b2476cf94f5df705e4f87146c18875c0af59741a&

is there a way to log semi-transparent Points3D or Elipsoids3D? I tried reducing the the alpha channel of the color, but that didn't work.

c++
rec.log(
    "sphere/point",
    rerun::Points3D({{sphere.center.x, sphere.center.y, sphere.center.z}})
        .with_radii({sphere.radius})
        .with_colors({rerun::Color(0, 0, 255, 5)})
);

c++
rec.log(
    "sphere/ellipsoid",
    rerun::Ellipsoids3D::from_centers_and_radii(
        {{sphere.center.x, sphere.center.y, sphere.center.z}},
        {sphere.radius}
    )
        .with_colors({rerun::Color(0, 0, 255, 5)})
);

https://cdn.discordapp.com/attachments/1379978709775286343/1379978711935357009/image.png?ex=685d3b93&is=685bea13&hm=1bd34107e68da97a68af404c17a0e601eabcf51bb7708b11358f59333491b0e9& https://cdn.discordapp.com/attachments/1379978709775286343/1379978712698851358/image.png?ex=685d3b93&is=685bea13&hm=50537b579e12a51f37dce13fb61642f92e48eaa05871de1749538023006c5ec5&

Hello, to be fair I am not sure whether this is an actual issue of just expected behavior but for my application (I am logging the current position of a humanoid robot) I need to log 33 Transform3D and 51 Scalars at minimum 10fps. What I am observing is that after around 2 minutes of logging the Viewer accumulate a certain delay showing old data instead of the most recently logged one. Is this expected? I would like to log rgb and depth and log at a higher frequency but that makes the issue much worse. I've attached an rrd to show the amount of entities I am logging. Rerun version: 0.23.2 Tested both on windows and linux https://cdn.discordapp.com/attachments/1379930097724690492/1379930098207031407/data.rrd?ex=685d0e4d&is=685bbccd&hm=3ac912c25f2fc84d595864be788aef2fe0c3c97812906d5e995dacd14526000d&

Hello, I want to use ReRun for AI robotics, to record datasets compatible with LeRobot models training. I assume I need to record my data with rerun, then implement a data pipeline to transform the data into the LeRobot format. Here are details about the LeRobot data specs: https://docs.phospho.ai/learn/lerobot-dataset Can you give me some pointers on how to do that? Would this be easy or would I end up reimplementing everything from scratch?

Hello! Is there a way to adjust the timeline size or hidden by default? I am rendering rerun visualizations in an iframe so the scene visualization ends up being really small https://cdn.discordapp.com/attachments/1379429260976062555/1379429261185912873/Screenshot_2025-06-03_at_2.57.32_PM.png?ex=6840351c&is=683ee39c&hm=45f7dee2cded1f814ad2c2c579593a7fbad60dcdbac9e76d825f5328bdf07174&

Hey everyone I'm new to using rerun, and Im unable to get the examples to run! I'm trying to use the 3d viewer, so I ran the below code in a jupyter notebook. """Log some very simple points.""" import rerun as rr rr.init("rerun_example_points3d", spawn=True) rr.log("points", rr.Points3D([[0, 0, 0], [1, 1, 1]])) rerun looks like its receiving the data, but the actual data field is empty. I'm not seeing any points popullating. Am I missing something?

Hello! I am trying to create a timeline to show past and future of some elements in a 3D scene. For these temporal elements I set rr.set_time_nanos("time", t) when I log them. However, all the static elements that should always be visible seem to get messed up when I do that even though I don't specify a time for them. Is the a way to log items and make them always visible no matter what timestamp I am on? I have my temporal elements under "world/dynamic" and the others "world/static" . I saw there is an "overrides" tried doing something there like: rrb.Vertical( rrb.Spatial3DView( name="3D View", origin="/world", time_ranges=[ rr.VisibleTimeRange( "time", start=rr.TimeRangeBoundary.cursor_relative(seconds=-10.0), end=rr.TimeRangeBoundary.cursor_relative(seconds=5.0), ) ], overrides={ "/world/static": [rc.Visible(True)], # Always visible }, but I get an error. I am a bit confused reading the documentation so any help would be appreciated 🙂

I'm migrating my rust application to 0.23 from 0.20. I see that

DisconnectedSpace

was deprecated, then removed. One migration document suggest that a deliberately invalid

Transform3d

can serve the same purpose, another document suggests the view-spawning heuristic was removed. So I just want to confirm, are Blueprints the only way to spawn views now? I don't have complicated layout needs beyond spawning views, so I'd rather not go the blueprints route while it's still python only. 2nd question: I see some blueprints items exported with the rust-sdk. Is it possible to do Blueprint stuff with the rust-sdk? I'm willing to work with an underdocumented API to avoid python, but I don't want to spend a lot of time on it only to eventually learn that's currently not possible. Thanks!

Hello. I noticed a strange thing. When I used

rerun.io/viewer

to view the results of data visualization using nuscenes, I noticed that the size of the loaded data was 103MB. When I saved it as an

.rrd

file, the size of this `.rrd

file was **93MB**. However, when I git the code to the local computer and ran the program, I found that the size of the loaded data was **396MB**. When I saved the data as an

.rrd` file, I found that the size of this

.rrd

file was 355MB. I'm very curious about why there is such a big difference in size between the data loaded in `rerun.io/viewer

and the data I load locally with code, and there is also some difference in size between the data saved in

.rrd` and the actual loaded data. I used the code implemented locally in python. Thanks! https://cdn.discordapp.com/attachments/1376846878133452801/1376847022241615882/data.jpg?ex=6836d036&is=68357eb6&hm=e1860eb9e02c092e68627fce6a78f37455c964878c1269c054a641440f49661e& https://cdn.discordapp.com/attachments/1376846878133452801/1376847022732087318/data_local.jpg?ex=6836d036&is=68357eb6&hm=348f94171b5e45e28853e5f8d28f04288294126f1f232b9cbc7b97d872577364& https://cdn.discordapp.com/attachments/1376846878133452801/1376847023071952896/data_rerun-io.jpg?ex=6836d036&is=68357eb6&hm=e7d11ac3be975f5c0810a8eae46f3adf7847efe2d312e1371eb0dbca2b970ab5&

I want to view logs live and also stream them to a file at the same time. My goal is to have live visualisations, but also store log files in case something goes wrong. My experience is that usually, combining spawn and save results in silent failures and nothing being logged. Right now, I'm at this piece of code

import rerun as rr

# Create a recording stream (acts as a log "sink")
rec_stream = rr.RecordingStream("my_app")

# Optionally, also spawn the viewer to see logs live
#rr.init("my_app", spawn=False)
rr.spawn(connect=True, recording=rec_stream)

# Also save the stream to a file
rec_stream.save("my_log.rrd")

# Now log your data
for i in range(1000):
    rec_stream.log("world", rr.Points3D(positions=[[1*i, 2, 3]]))

which does not log to the saved file. Changing the order of lines changes the behaviour: sometimes I get logs in the file, sometimes in the live viewer, but never in both.

I'm looking into using wgpu code through re_renderer, but the different caches and handles are forcing me to rewrite most of the code with a bunch of rerun specific boilerplate to fetch cache (I have to use hashable wrappers and their keys), it would come a long way if I could just reuse a wgpu struct and hide the cache logic. maybe a "simple" transformation function could be used: I suspect it would greatly reduce the mental overhead of renderers too.

rs
impl RenderPipelineDesc {
    pub fn try_from(
        ctx: &RenderContext,
        // The user doesn't need much rerun knowledge, plain wgpu is enough.
        desc: wgpu::RenderPipelineDescriptor<'_>,
    ) -> Result<RenderPipelineDesc, ()> {
        Ok(Self {
            label: desc.label.into(),
            // cache key is retrieved automatically
            pipeline_layout: ctx.gpu_resources.pipeline_layouts.get_or_create(
                ctx,
                // Cache again
                PipelineLayoutDesc::try_from(ctx, desc.layout.ok_or(())?)?,
            ),
            // etc...
            ...
        })
    }
}

posting as discussion to gauge if that's worth an issue.

I'm looking into using rerun stack for visualizing real-time fluid simulation: https://github.com/Vrixyz/rerun/pull/2 ; I'd like to render my GPU data, can you help me through it ? How it works currently: - I use re_renderer - each frame, I run my own compute shader to advance the simulation - each frame, I read GPU data into CPU to write it back to GPU, through

PointCloudBuilder

. - I'd like this step to stay in GPU, likely a custom shader step to copy my own data into the data for rerun's point cloud ? Does that make sense ? I'm not too sure where to start, and I'm not sure how to sync the batch data 🤔.

Hello, I really like the light theme mode you added in rerun 0.24.0a1. However, I want rerun to display the light theme directly through code. I'm using python-sdk. Besides, for the time being, I can only find the Rerun Python APIs of rerun rerun 0.23.2. I hope you can provide the rerun Python APIs of rerun 0.24.0a1. Looking forward to the official release of your version 0.24.1. Thanks!

Hello, is there a way to create a blueprint object in code from an rbl file? I'd like to load a blueprint saved in an rbl file when spawning a rerun viewer in code. I'm using the python sdk.

Hey everyone, We've got a critical performance issue with the rerun web viewer. It's trying to load multiple large videos, resulting in the rrd file being larger than 10GB+ file being pulled into memory. This makes the viewer incredibly slow, buggy, and often causes crashes. Data streaming is absolutely essential here to make the viewer usable. My Questions: When can we expect this streaming feature to be ready? Can we prioritize this development? The current state is a major bottleneck. Any insights or experience with implementing video streaming in web viewers would be hugely appreciated! Thanks!

Using

0.22.0

, I am logging lots and lots of

rr.Mesh3D

. When looking at the

rr.SpatialView3D

, none of the meshes are visible. If I click through them, and open up the entity in the tree, I see that no Visualizers are selected. If I manually add a Mesh3D visualizer, it shows up. (other options are rr.Arrow3D and rr.Points3D. Is there any way I can make it the default? I couldn't really figure out the overrides regarding visualizers yet. Note, that I have hundreds of meshes in a deeply nested tree, so manually clicking through it would be a pain.

How to set duration if I want some geopoints to stay longer than just the setTime, the way I am getting past this right now is setting a new entityPath for each drawing, even though it should be the same and it slows down the application quite a bit doing it this way.

When I log a lot of data to rerun (a few hundred entities every 50ms), the visualizer embedded in gradio becomes laggy and some logged data is lost. If I visualize the data just with rerun (with no gradio) everything works perfectly.

I am having issues where my timeline is not showing up. Using 23.2 and linux, I am sending in two timestamps using

dt = datetime.datetime.fromtimestamp(t_ns / 1e9, tz=datetime.timezone.utc)
                rr.set_time("timestamp_gimbaled_camera", timestamp=dt)

dt = datetime.datetime.fromtimestamp(int(ts_ns) / 1e9, tz=datetime.timezone.utc)
                rr.set_time("timestamp_vehicle", timestamp=dt)

https://cdn.discordapp.com/attachments/1374480038803800154/1374480039038550156/image.png?ex=682e33c9&is=682ce249&hm=e0812f7e54f033dbebbfc109b6f11074ec0ef93bbf3b439dce29dcfbb69bf9ee&

Hi! We have a use case where we want to log information from various sources that use different timestamps (ticks vs actual time). We have at least one source that contains values for both timeline. Right now we're using a manual asOf join to put everything onto the same timeline before logging, but can't rerun also take care of that for me? I feel like that would be one of the main reasons why someone would use multiple different timelines in Rerun, but I can't find anything about that documented.

Hello, I want my view to display the entities (they are scalars)

poses/action/components/x
poses/proprio/components/x

and right now I'm using the filters

+ poses/action/components/x/**
+ poses/proprio/components/x/**

I would like to have a more general filter like

+ poses/**/x/**

I know I could restructure the entity paths to end with the values I want to generalize on e.g.

+ poses/x/action
+ poses/x/proprio

but that doesn't fit my visualizer logic.

Hi All, I'm trying to visualize some simple 2D data in rerun, but nothing is showing up inside the main viewport, even though if I check the data in rerun, it all looks correct, except for one thing: All values in my "visual bounds 2D" are NaN. My guess is that these visual bounds are the reason why I'm not seeing anything, but I can't figure out why this is happening. Any idea what I'm doing wrong? A minimal example with which it happens is this

auto rerun_recording_stream = std::make_unique<rerun::RecordingStream>("Navigation");
    rerun_recording_stream->connect_tcp(fmt::format("{}:{}", rerun_ip, rerun_port)).exit_on_failure();
    rerun_recording_stream->log("path", rerun::LineStrips2D{ rerun::LineStrip2D{ {
                                            { 1.55f, 3.03f },
                                            { 11.27f, 6.05f },
                                            { 8.07f, 14.41f },
                                            { 15.25f, 23.05f },
                                        } } });