Tilebox retweeted
You really want workflow changes to stay independent of the infra team
And now we're there finally: You set up Tilebox runners once – for now ;) – then manage config / settings and secrets in Tilebox, with shared defaults and overrides. Same workflow for prod / staging or different customers, with full transparency and no infra redeploys. API key rotation is trivial and automated too, now for even better security.
tilebox.com/dispatch/news/en…
Tilebox retweeted
You can now tag and filter imagery in Tilebox.
I’m especially excited about semantic labels. Tag what’s in an image, so downstream can filter before they look at it.
Enums give you a defined vocabulary across your index and deterministic filtering.
tilebox.com/dispatch/news/da…
Tilebox retweeted
We're building a hub of EO algorithms that can be one-click deployed, with open source workflows you can use as templates and adapt
Which geospatial workflow would you or your customers want to run first? What makes it difficult today?
If you build algorithms, what would you need to make yours available through something like this? @WyvernSpace @Esri @SkyWatchApps @NASALifelines @CloudFerro
Tilebox retweeted
Tilebox now supports workflows triggered by local file arrivals!
Means you can process e.g. drone imagery locally or on your workstation, or test event-driven workflows locally with the same task interface used for cloud storage for whenever you want to move to the cloud
Full drone workflow comin soon!
tilebox.com/dispatch/news/lo…
Tilebox retweeted
About 40% of our users are on Windows, and not all of them use WSL
So we’ve added native Windows support to Tilebox: Friends on Windows, you can now natively develop and test workflows locally on your own machine, then deploy them as usual to Linux for prod
tilebox.com/dispatch/news/na…
Tilebox retweeted
When satellite imagery arrives in your bucket, you want to process it right away and keep improving the algorithms without touching infrastructure.
That’s a lot easier now with storage automations in Tilebox. You can even iterate on your laptop!
tilebox.com/dispatch/news/cl…
Tilebox retweeted
Earth data processing pipelines typically start with a few tasks and then, rather quickly, grow into hundreds or thousands
So we redesigned the job view to make those larger jobs easier to understand. Tasks now appear in their workflow hierarchy, so you can see them in their natural context.
Expand a branch, select a task, and inspect its input, timing, retries, compute location, trace, and logs.
Super quick to navigate too
Check it out console.tilebox.com/home/qui…
Tilebox retweeted
Where to go to see the 12 August total solar eclipse?
Inland northern Spain!
Best locations:
• Valladolid: 0.0% forecast cloud, Sun 8.5° above the horizon
• Burgos: 0.9% cloud, Sun 8.0° high
• León: 2.7% cloud, Sun 9.5° high
• Santander: 15.8% cloud, Sun 8.8° high
Maximum eclipse is around 20:30 local time. Totality happens close to sunset.
London and Paris are great too:
• London: 91.4% obscuration around 19:13, Sun 10.3° high
• Paris: 92.1% around 20:18, Sun 7.5° high
Both with 0% cloud forecast. Partial eclipses though
Reykjavík gets totality but is completely cloud-covered.
Contributions: ECMWF for the weather, JPL/Skyfield for the astronomy and Tilebox turned this into a proper workflow.
Here's the map:
storage.googleapis.com/tileb…
Tilebox retweeted
Tried this across an entire German state
Three prompts and Tilebox ran zero-shot Qwen3.8-Max across Brandenburg. We found 247 PV power plant candidates.
82 km² at high confidence, official capacity implies ~87 km².
Almost the whole expected footprint in one shot!
Tilebox retweeted
Geospatial pipelines have a habit of turning into infrastructure projects.
I wanted to see how fast I could go from an empty folder to a deployed Tilebox workflow, so I recorded it.
90 seconds: start the runner, initialize the Python project, publish, submit a job, then check the logs and trace.
The nonstandard part is that the runner can start before the workflow exists. It can stay on your cloud, laptop, or on-prem machine while new workflow releases come to it.
This one says hello world, and in the next one we'll do some actual work.
Tilebox retweeted
Want to vibe with SPACE DATA? Now you can. Claude + Tilebox = any satellite data, any insight, one prompt. 🛰️
In the last five years we sent more satellites into space than in the 50 years before that. China alone requested licenses for 200,000 satellites this year. SpaceX asked for a million. For context: right now there are 15,000 active ones up there.
Space is getting crowded. And it's producing insane amounts of data. The problem was never the data, the problem was using it.
Until now, satellite data was for hedge funds monitoring Walmart parking lots. You needed massive teams, massive infrastructure, and 80-90% of the work was just data cleanup and sourcing before any actual analysis happened.
Tilebox just changed that. Their pitch: the AWS for SPACE DATA.
You connect Claude Code to their MCP, write a prompt, and it figures out the rest. You don't need to know where the data is stored, how it's formatted, or which AI model to use. It finds the data, runs the compute co-located with it, and gives you answers.
And this is just day 0 - imagine how this will look once there are 100 thousands of satellites in space (we got 15k now). This is exactly the kind of software infrastructure Europe should be building. 🇪🇺
Disclaimer: we're investors (obviously). Also means we got first dips in showing this to the world.
Tilebox retweeted
A lot of satellite and geospatial work gets stuck between “the analysis works” and “someone can use this reliably.”
We’re testing a Tilebox implementation sprint to help close that gap.
Bring us an algorithm or workflow that already works in a notebook, script, or internal pipeline. We’ll help implement it in Tilebox and turn it into a repeatable workflow that can be run, monitored, debugged and operationalized.
We’re looking for 3 to 5 teams to try this with. DM me if you have something in mind!
Tilebox retweeted
If you're curious, ask your favorite model to research for you how much CO2 is emitted by transferring 1 PB of data between cloud regions, vs the CO2 emissions of an afternoon of agentic coding!
Surprised me quite a bit, it's a factor 2,000x to 400,000x The absolutes (in driven mile equivalents for example) are shocking
Better to invest a few tokens and an afternoon and co-locate your algorithms with the underlying data, now that the infra difficulties are removed
tilebox.com/blog/climate-tec…
Tilebox retweeted
Tilebox 2.0: Earth data for wherever your agents work
producthunt.com/products/til…
Upvote if you'd like to support us and make physical world data more accessible and actionable!
Tilebox retweeted
Physical-world data has reached a strange point:
we can capture it faster than most teams can use it.
The data is still operationally far away from the software trying to use it.
Tilebox gives teams the missing loop: discover data, run workflows where the data lives, observe what happened, debug, and rerun.
Tilebox retweeted
.@tileboxio announced an update aiming to make AI agents more effective geospatial-data analysts.
payloadspace.com/exclusive-t…
Tilebox retweeted
As part of the launch we built datacenterbuildout.com with an agent on Tilebox.
The site tracks US data center construction using satellite imagery, with a web frontend, backend workflow, before/after imagery, rankings..
Here's what we didn't have to do:
- set up workflow infrastructure
- build a complex CI/CD pipeline to roll out changes
- wire up observability
- copy-paste logs, traces, errors, and failure modes back into the agent
- explain where the satellite data lives or how to load it
- tell the agent how to parallelize the work
- manually deploy each workflow iteration
That's the development loop that this launch unlocked
data context → workflow code → deployment → observable jobs → inspectable outputs → iteration
Tilebox retweeted
Today we launched a major Tilebox update, covered exclusively by Payload.
The core idea is: AI agents should not just answer questions about geospatial data. They should be able to build, run, inspect, and improve the workflows that produce those answers.
For physical-world data, the hard part is not only the model. The data is huge, fragmented, governed, and spread across providers, clouds, internal systems, sovereign environments, and eventually orbit.
If an agent has to guess the dataset, schema, runner, logs, or output provenance, it will eventually make things up.
So we built Tilebox around the engineering loop: discover typed data, define a workflow, run it where the data lives, observe logs/traces/task status/outputs, debug, and rerun.
That is what makes agents useful for EO: reviewable workflow execution, not black-box answers.
Thanks to @payloadspace for covering the launch.
payloadspace.com/exclusive-t…
Tilebox retweeted
We're excited to sponsor Berlin Aerospace Hack, and I'm personally excited to be there at the kick-off tomorrow!
To offer the most, we're boosting @tileboxio Community Access tier capacity for this weekend, and can't wait to see what Berlin will build
140 people, one day and two nights
Let's go 🚀
Tilebox retweeted
🚀 Hackathon Challenge #3: Mini-Constellation Tasking Planner
Build a 24–72h imaging plan for a 3–6 sat constellation: maximize AOI coverage, avoid clouds, respect sun-angle + off-nadir limits.
This is real mission-planning work.
Use @tileboxio to:
🛰️ Query ephemerides, terrain masks, cloud forecasts
⚙️ Run batch schedule evaluations
💾 Reuse cached visibility + weather lookups
🌍 Pull open EO data (S1/S2/S3, Landsat)
Deliver:
• Coverage map (footprints by sat/time)
• Tasking schedule (CSV/JSON)
• KPIs: coverage %, off-nadir, cloud-free area
• Compare vs. naïve baseline
This is exactly how real operators plan imaging windows.
Join the Berlin Aerospace Hackathon ↓
luma.com/scw0gyac



















