Comparison · Robot data tools
Coldstack vs Foxglove
Short version: they solve different problems, and they work well together. Foxglove visualizes and debugs a recording. Coldstack searches a whole fleet of recordings to find the ones worth opening.
Two different jobs
Foxglove is a robotics observability platform — a mature tool for visualizing, debugging, and collaborating on robot data. You open a recording and see exactly what happened, panel by panel. It is excellent at that, and it created MCAP, the open log format the field now standardizes on.
Coldstack is a retrieval layer. Before you can inspect a moment, you have to find it — and across a fleet of thousands of recordings and petabytes of data, that is the hard part. Coldstack indexes the whole fleet on object storage and lets you search it by metadata, signal, and image or text meaning, then export the matches as training data.
Side by side
| Foxglove | Coldstack | |
|---|---|---|
| Primary job | Visualize and debug robot data | Search a whole fleet's logs |
| Best at | Opening a recording and seeing what happened | Finding which recordings and moments matter |
| Scale | One recording at a time | The entire fleet corpus at once |
| How you find data | Open a file, scrub the timeline | Query by metadata, signal, and image/text meaning |
| Where your data lives | Uploaded to / managed by the platform | Stays in your own S3 bucket; index-only custody |
| Main output | Panels, layouts, dashboards, replay | Ranked moments + export to LeRobot training data |
| Log format | Created MCAP; native support | Indexes MCAP directly |
They're complementary
The two fit together: use Coldstack to search the fleet and find the moment, then open it in Foxglove to inspect it frame by frame. Coldstack even ships a Foxglove extension — run a search and jump straight to the matching episode in the timeline. Both build on MCAP, so there is no format friction.
A simple way to hold it: Foxglove answers "what happened in this recording?" Coldstack answers "where, across everything, did this happen?" — and turns the answer into a training set.
Which should you use?
- Inspecting or debugging a recording, building dashboards, live monitoring → Foxglove.
- Searching a whole fleet for failures or edge cases, curating training data → Coldstack.
- Both → use Coldstack to find it, Foxglove to look at it.
Questions
Is Coldstack a Foxglove alternative?
Not exactly — they do different jobs. Foxglove is best-in-class for visualizing, debugging, and collaborating on robot recordings: you open a file and see what happened. Coldstack is a search layer: it finds which recordings and moments matter across an entire fleet, then exports them as training data. If your problem is 'inspect this recording,' that's Foxglove. If it's 'find every recording where this happened,' that's Coldstack.
Do Coldstack and Foxglove work together?
Yes. They are complementary: use Coldstack to search the fleet and find the moment, then open that moment in Foxglove to inspect it. Coldstack ships a Foxglove extension that lets you run a search and jump straight to the matching episode in the timeline. Both build on MCAP, the open log format Foxglove created.
What does Foxglove do that Coldstack doesn't?
Foxglove is a mature visualization and observability platform — rich panels, custom layouts, live streaming, and team collaboration for inspecting and debugging robot data. Coldstack does not try to be a viewer; it is the retrieval layer that finds what to look at. For inspection and debugging, Foxglove is the tool.
What does Coldstack do that Foxglove doesn't?
Coldstack searches across an entire fleet's logs at once — by metadata, by time-series signal, and by semantic image or text match — and exports the matches as a LeRobot-compatible training dataset. It is built natively on object storage, so the raw logs stay in your own bucket and a query costs a bounded read rather than an always-on database.