Two switches decide how much of your working day your colleagues can read. They are the most consequential controls in the product, so this page covers what each one turns off, including the things it turns off downstream. That is the part people are surprised by.
Both live in Settings → Sharing.
Switch one: What I ask#
| State | What a teammate sees |
|---|---|
| On | Ravjeet prompted Claude Code, followed by what you wrote. |
| Off | Ravjeet prompted Claude Code, with the prompt masked. They know you are working. They cannot read the words. |
Switch two: What my agent changed#
| State | What a teammate sees |
|---|---|
| On | The lines your agent changed, and the agent's visible reply for a turn. |
| Off | Which files moved, and nothing about their contents. The detail is still written on your own machine, so your sessions read in full for you. |
What is shared in every mode#
Which files you are touching. Always, including in the most private configuration.
That is the trade at the centre of the product. Your prompts can stay entirely on your machine, but if nobody can see which files are moving, nobody can avoid you, and avoiding each other was the point. Your choice of mode is itself shown to teammates, so nobody is ever watched without knowing.
Secrets are stripped on your machine before anything is sent, in every mode.
The downstream effects nobody expects#
These switches gate more than the feed, because several features are built out of the material the switches control. Runi turns a feature off rather than publishing something you did not agree to.
| Turning off… | Also stops | Because |
|---|---|---|
| What I ask | Task cards being rewritten from your agent's reply | Somebody who hides their prompt has not agreed to publish their agent's restatement of it. |
| Model-written task briefs | The same reason. Cards keep the local brief, written from your prompt on your machine. | |
| What my agent changed | Decisions, discoveries and blockers being extracted | They are read out of the agent's visible reply, which is what this switch governs. |
| The reply-based task intent | Same source. |
Tip
Memory is the exception. Asking your agent to remember something for the team still shares the memory, even with detail kept local, because that request is the same act as typing it into Memory yourself. The backing event stays on your machine; the entry travels.
A third, smaller switch#
Settings → Data & privacy carries Share test commands, which decides whether a test event may carry the command that produced it. Pass and fail counts publish either way, because that is how a teammate learns something broke.
Two things on the same page are guarantees rather than settings, because they cannot be switched off:
- Whole files are never uploaded. Only the lines an agent changed, and only when detail sharing is on.
- Secret detection is always on. Every event is scanned for keys, tokens and credentials before publishing, and matches are masked on your machine.
The same page has a redaction playground: paste something in and see what would be masked, without publishing anything.
See it as a teammate would, before you decide#
The settings page previews your own most recent real prompt, rendered the way a colleague would receive it under each setting. Seeing your own words masked is the only version of this that answers the question you are asking when you open the page.
If you have never prompted, it shows a clearly-labelled example instead. Runi will not invent a plausible-looking prompt of yours to fill the space.
Choose deliberately, as a team#
The failure mode here is not a wrong setting. It is a default that wins by inertia because nobody could tell from three words what a teammate would end up seeing. Spend five minutes on it in week one.
| If your team… | A reasonable default |
|---|---|
| Works on one product, trusts each other, wants maximum value | Both switches on. Everything downstream works, and briefs are at their best. |
| Handles regulated or client data in prompts | What I ask off, what my agent changed on. Files, changes and extracted decisions still travel. |
| Is evaluating Runi and wants the smallest footprint | Both off. You still get overlap warnings, file activity, task cards from local briefs, and memory you write yourself. |
Changing them needs the desktop app#
These are enforced on the machine your agents run on, so the browser build shows you what is in force but will not let you change it. Open Runi on the machine itself.
They are per person, not per workspace#
Nobody can set your switches for you, including an owner. What an admin can do is see which mode you have chosen, which is the honest half of the bargain.