Made for Swiss self-employed & SMEs
An AI assistant usually waits for a question. OpenDots aims to be more like a named coworker: it can remember previous work, use a browser or files, and carry out a task on a computer. But OpenDots is an early open-source project, not a ready-made office service. The sensible question is not whether it can do everything; it is whether one small, supervised task is worth testing.
What OpenDots actually does
OpenDots lets you create a persistent AI teammate with a role. it can keep context between sessions. The project describes ways to talk to it by text or voice, and to connect it to Slack. Its computer can open a browser, work with files and run commands to complete a task.
A teammate might gather details from an approved folder, prepare a first draft, or organise information for a person to review. The project also describes approval pauses before actions such as saving or sending something. Those are product claims from the project, not an independent safety certification.
OpenDots is an MIT-licensed template that you host yourself. It does not include its own AI model: you must connect a compatible model service, and Slack use needs Slack credentials. Hosting the application yourself does not automatically mean the model, logs or every piece of data stay on your own computer.
Where it might help a small office
Start with work that is repetitive, easy to check and reversible. For example, an assistant could prepare a daily list of unanswered enquiries from a folder you have chosen, or draft a polite reply using a short set of approved answers. A person should check names, dates, promises and attachments before anything leaves the office.
Do not begin by giving it access to every mailbox, client folder or calendar. Choose one low-risk process and write down what a good result looks like. For a first trial, keep the agent in draft mode: it may prepare work, but a human performs the final send, booking or record change.
- Pick one task that currently takes a few minutes each day.
- Give access only to the files and accounts needed for that task.
- Keep a person responsible for checking and approving the result.
Approval is useful, but not the whole security plan
A confirmation button can prevent an unwanted message from being sent. It cannot by itself prevent the agent from seeing more information than it should, making a misleading draft, or passing client details to the model service. Nor does a pause prove that every action is logged or can be undone.
Before using real business information, find out who operates the server and model, where information is processed, how long prompts and logs are retained, and who can access them. Check the project maturity too: its own description calls it alpha, says it is for a single owner, and notes that shared editing is not yet available. Voice and Slack delegation still need more real-world verification.
For a Swiss practice handling health, legal or financial information, use synthetic examples first. Ask your IT provider or data-protection adviser to review the setup and the relevant service terms. If you cannot explain where a client message goes, do not connect that mailbox yet.
A sensible way to decide
OpenDots is worth watching if you have technical support and want to experiment with a supervised digital helper. It is not the quick, plug-in solution its coworker metaphor might suggest: someone must host it, connect a model, set permissions and keep it updated. A paid hosted assistant may be simpler,
Try a small, non-sensitive pilot for a week. Track minutes saved, corrections needed and any moments when the agent asked for more access than expected. Stop if the human review takes as long as doing the task yourself, or if the data path is unclear. Keep the useful idea — a helper that prepares work and waits for approval — even if OpenDots itself is not yet right for your office.
