River
Ctrl K
Open River

River AI

Edit, Plan & Ask

SpaceChat exposes three modes to control how much the AI can act. Each one offers a clear guarantee about what will and won't happen.

Ask mode#

Use Ask when you want to think out loud. The AI can read your Dataspace, reason over it, and answer questions — but it does not mutate state.

Guarantee: Nothing changes. No entities are created. No emails are sent. No tools that have side effects will run.

Use Ask for:

  • Exploring options before deciding
  • Asking "what is in my workspace?"
  • Getting a second opinion on a draft
  • Understanding what an Agent would do without actually running it

Plan mode#

Use Plan when you want a structured action plan before anything happens. The AI researches, drafts a plan, and stops — waiting for you to approve or edit before execution.

Guarantee: The AI writes a plan. It doesn't act on it.

Use Plan for:

  • Multi-step tasks where you want to see the steps first
  • Risky actions you want to inspect before running
  • Complex changes across many entities

When you're satisfied with the plan, switch to Edit and ask the AI to execute it.

Edit mode#

Use Edit when you're ready to ship. The AI can read context, call tools, write entities, send messages, and otherwise act in your workspace.

Guarantee: The AI does the work. You see what it did in the chat thread and in Activity.

Use Edit for:

  • Routine work you've done a hundred times
  • Anything where the right outcome is well-defined
  • Following up on a Plan you've already approved

How to choose, fast#

  • Still exploring? Ask.
  • Want to see the plan first? Plan.
  • Ready to ship? Edit.

This three-mode pattern is how you stay intentional about AI autonomy without giving up its speed.

Switching mid-conversation#

You can flip between modes inside a single thread. A common pattern:

  1. Ask — "what's the state of the launch doc?"
  2. Plan — "what would I need to update to ship Friday?"
  3. Edit — "do the things in that plan."

The AI keeps the conversation context across mode switches. You're just changing the rules of engagement.

Tips#

  • The mode is per-Space, not per-message. The selector is in the SpaceChat input bar.
  • A confident "this should be safe" hint from the AI is not a substitute for Plan mode. Use Plan when the action is risky, even if the AI sounds sure.
  • When in doubt, use Plan. It's cheap to read; mistakes in Edit can be expensive.