What this article covers
A guide for product teams, from ZumiLabs. What "local-first" means here.
When a team picks a tool, the comparison usually comes down to features and price per seat. Cloud tools win that comparison easily: nothing to install, works everywhere, starts free. The costs that tip the balance don't appear on the pricing page. They show up later, in three forms: data, latency and lock-in.
This isn't an argument against the cloud. Plenty of work belongs there. It's a list of what to weigh, so the choice is made on purpose.
Data: copies you can't count
Upload a file to a cloud tool and it rarely exists in just one place. There may be the stored original, processed versions, previews, search indexes, backups, logs, and copies held by the services the tool itself relies on. Each copy has its own retention period, and deleting the original doesn't necessarily delete the rest.
For most files that's fine. For some it isn't: customer recordings, unreleased designs, health information, anything covered by a contract or a law such as GDPR. Then someone has to answer questions like these:
- Which countries is the data stored and processed in?
- Which other companies process it on the tool's behalf?
- Is it used to train or improve anyone's models, and can that be switched off?
- When we delete it, what is actually deleted, and when?
Every one of those questions has an owner, a document and a review cycle. A tool that never receives the data skips the whole list.
Latency: waiting on the round trip
A cloud tool has to move your work to its servers before it can do anything. For small things that's instant. For media it isn't. A 2 GB screen recording on a 20 Mbit/s upload takes about 13 minutes to arrive, before any processing starts. On a train, in a hotel or on a busy office network, it may not arrive at all.
Waiting also has costs beyond the minutes:
- Lost focus. A task that takes a few seconds keeps you in the flow. One that takes several minutes sends you to email, and you come back later, cold.
- Fewer attempts. When every try means an upload and a wait, people try once and accept the result.
- Offline means stopped. No connection, no tool.
Work done on the machine where the file already is skips the round trip. That's how we transcribed 19 hours of course audio in 14 minutes without uploading a single file.
Lock-in: work that only opens in one place
Lock-in rarely looks like a trap. It looks like convenience: a format only that tool reads, links that only work while you subscribe, history and settings that don't export. The cost appears when something changes:
- Prices go up, or the free plan shrinks, and moving out is harder than paying.
- The product changes direction, or is acquired, or shuts down, with a deadline for exporting your data.
- Your needs change, and the work you made is stuck in a tool that no longer fits.
The test is simple: if this service disappeared tomorrow, what would you still have, and could you open it?
Where the cloud is the right answer
None of this means "never use the cloud". It is clearly better for:
- Real-time collaboration, when several people edit the same thing at once.
- Heavy computation that an ordinary laptop can't do well, such as the largest AI models or long video generation.
- Sharing and publishing, when the point is for other people to see the work.
- Teams without anyone to manage devices, where one well-run service is safer than fifty unmanaged laptops.
The good pattern is to keep the work local by default and send things to the cloud deliberately, one job or one file at a time, when the cloud clearly does it better.
Questions to ask before choosing
Before your team's files go into a new tool, ask:
- Does the work need to leave our machines at all? If not, prefer a tool that doesn't send it.
- Who else receives it? Ask for the list of sub-processors, not a summary.
- Is it used for training? Get the answer in writing, with the setting that controls it.
- How long does a typical job wait on upload, for our real file sizes and our real connection?
- Does it work offline, or degrade gracefully?
- Can we export everything in a format another tool can open, including history and settings?
- Can we bring our own keys, so AI and storage costs are billed to us directly, with no markup?
A tool that answers all seven well might still be a cloud tool. But you'll have chosen it knowing the full cost, not just the price.
Tools that keep the work where it is
Browser tools and Mac apps that run on your own computer by default, with cloud services only when you choose them, using your own keys.
Explore the tools