At a glance
| Public servers | Your own | Paid (Overspan) | |
|---|---|---|---|
| Cost | Free (donated) | Server bill plus your time | $19 to $139 a month |
| Volume you can plan around | ~100 queries and 10 MB a day for regular use | Whatever your hardware handles | 50,000 to 1,000,000 requests a month |
| Commercial use | Asked to go elsewhere | Yes | Yes |
| Works from cloud IPs and CI | Not reliably | Yes | Yes, the key is the identity |
| Historical (attic) queries | Yes, on the main instance | If you enable it | Not yet |
| Setup time | None | Days, plus ongoing upkeep | Minutes |
| When it breaks | Wait, or try another instance | You fix it | We fix it |
The public servers: a gift, with rules that come with it
The main public instance at overpass-api.de is run by FOSSGIS, a
German non-profit, on donated resources. A handful of other organisations run
free instances too. They have carried most of the world's Overpass traffic for
years, for no money, and almost everyone who uses Overpass started there. We did.
The rules are written down, and they are reasonable. The main instance's usage policy on the OSM wiki says you will not disturb anyone below 10,000 queries and 1 GB of downloads a day for one-off use. For anything that runs regularly, divide both numbers by 100: about 100 queries and 10 MB a day. If you run an app or website, that budget covers all your users put together, not each of them.
The same policy is direct about who it is not for: "Commercial use should use self-hosted or paid Overpass servers." It also asks for no parallel scripts, a User-Agent that identifies your app, and a 30-second pause after any 429.
What you get in return is real. Full planet, minutely updates, full metadata,
and attic data on the main instance, so historical [date:] queries
work. That last one is something most paid instances, ours included, do not
offer yet.
What you do not get is a promise. The wiki's own status note for the main instance says it is overloaded and asks users not to expect high reliability. The Overpass manual explains how the server protects itself: requests wait up to 15 seconds for a slot, then get a 429, and when load climbs, shedding starts with the heaviest users first. Since 2026 the operators have also blocked stock client User-Agents, parts of the AWS and Azure ranges, and some app-hosting platforms, because of abuse from those sources. We covered the mechanics in the 429 post and the cloud-IP post.
None of that is a complaint. A free service run by volunteers has to defend itself, and those defences are exactly why it is still up. The point is that it was built for people exploring the map, not for your production job.
The public servers are the right choice when: you are prototyping, writing queries in overpass turbo, doing a one-off extract, running research at human pace, or you stay comfortably under the regular-use guidance.
Your own instance: full control, and a second job
The Overpass software is open source, and running your own copy is a well-trodden path. It is the same software the public servers run, so your queries do not change. There are no caps but your hardware, you can turn on attic data if you need history, and at high enough volume it is cheaper than paying anyone. The Overpass manual itself lists the jobs that belong on your own instance: sending the same request tens of thousands of times a day, fetching elements one by one millions of times, stitching bounding boxes to scrape the whole world, and backing an app for people who are not mappers.
What you take on, honestly:
- Disk. A full-planet copy with metadata lands near half a terabyte and grows every year.
- Memory. Enough RAM that queries are not bound by the disk, which is the difference between fast and unusable.
- Time to first query. The initial import and catch-up take a day or more before it serves anything.
- The update loop. It has to be watched. Its usual failure is to stop quietly while the server keeps answering every query, perfectly, from data that is getting older by the hour.
- Everything else a server needs. Security updates, disk alerts, restarts, and someone who knows what to do at 3am when it falls over.
None of that is exotic. It is also not free, even when the server bill is small, because the real cost is the hours. If you already run infrastructure and your volume is large and steady, this is often the right call, and the public servers' operators will thank you.
Your own instance is the right choice when: your volume is large or constant, you need attic data or custom settings, you already have someone who looks after servers, or the numbers work out cheaper than any paid plan.
A paid instance: the middle
Between the free pool and your own server sits a paid, hosted instance. Several exist, listed alongside the public ones in the OSM wiki's instances table. We run one, Overspan, so read this section knowing that.
What a paid instance changes is the shape of the deal. Instead of a shared queue and guidance you hope you are inside, you get a key and written limits. Overspan's smallest plan is 50,000 requests a month, about 1,600 a day, or roughly sixteen times the public regular-use guidance. The key is the identity, so the same script runs from a laptop, a Lambda, or a CI runner without a User-Agent check or an IP block in the way. Every response tells you how much quota is left, and hitting a cap gets you a clear error rather than a surprise bill.
It is the same Overpass software again, with the full planet, minutely updates and full metadata, so moving is one URL. Moving away is also one URL, which is why we publish a guide to leaving.
Where it is not the answer, honestly:
- It costs money. There is no free tier and no trial. If the public servers already cover your job, keep using them.
- No attic data yet. If you need historical queries, the main public instance or your own server does that today.
- Caps are caps. Each plan has a per-query timeout and memory ceiling. A country-scale extract needs the larger plans or a query split into pieces.
- At very high, steady volume, running your own wins on cost. We will say so if you ask.
A paid instance is the right choice when: you have a scheduled or production job that has outgrown the public guidance, it runs from the cloud or CI, and you would rather not maintain a server to keep it running.
A quick way to decide
- Under about 100 queries a day, non-commercial? Use the public servers, and follow their rules.
- Commercial, or regularly over that? You are past what the public servers ask for. Pick between the next two.
- Do you have server time to spare, or huge steady volume? Run your own.
- Otherwise? Pay for an instance, and put the hours into your product instead.
Whichever you pick, set [timeout:] and [maxsize:]
explicitly in every query. Your queries then behave the same on any Overpass
server, and switching between these three options stays a one-line change.
Look after the commons
Every Overpass instance, paid or free, serves data that OpenStreetMap volunteers mapped and that FOSSGIS and others keep flowing. Moving a heavy job off the public servers is one of the most useful things you can do for them. If they have saved you time, consider donating to OSM or to the people who run the instance you use.
Sources: the main instance's usage policy and status note are quoted from the OSM wiki's instances table as read on 2026-09-24; server capacity, queueing and the run-your-own list come from the commons chapter of the Overpass user's manual; Overspan's limits are from the pricing page. If anything here goes out of date, mail hello@overspan.dev and we will fix the post.