River AI
Overview
River AI is not a single chat box bolted onto a notes app. It's a layered intelligence stack with multiple modes of action — each one suited to a different kind of work.
The key idea: in River, the AI doesn't wait for you to invoke it. It's already inside the system, watching, learning, and acting on your behalf when relevant.
Four modes of AI#
Reactive — SpaceChat#
You ask, River answers. SpaceChat is the AI sidebar in every Space and the fastest way to do almost anything. It can read your Dataspace, take actions, run tools, and write entities — and it has three modes (Edit, Plan, and Ask) so you control how much autonomy it has.
Scheduled — Agents on cadence#
Some work is predictable. Your Morning Briefer runs at 7am. Your Weekly Recap runs at 4pm Friday. Your monthly close runs on the first. Agents + Schedules turn recurring work into infrastructure.
Triggered — Agents on events#
Some work should happen the moment something changes. An email arrives from a known investor → your Deal Researcher wakes up. A new ticket lands → your Support Triage Agent drafts a reply. A transcript finishes processing → your Standup Synthesizer turns it into action items.
This is the "ambient" mode that makes River feel like a coworker. See Triggers.
Ambient — Signal Intelligence#
A background system that scores everything in your Dataspace for relevance and surfaces what matters before you ask. A board meeting tomorrow with a thread that's gone silent. A stalled deal nobody's followed up on. A pattern across your inbox you wouldn't have noticed.
See Signal Intelligence.
The execution chain#
Whichever mode wakes the AI, the same chain runs underneath:
- A request, event, or schedule fires.
- River chooses a model (cloud or local) via the Intelligence Router.
- An Agent, Workflow, Skill, or tool run starts as a Job.
- Work executes — calling tools, reading entities, writing new ones.
- Output lands as entities in your Dataspace.
- Notifications and surfaces reflect the result.
Every step is recorded. Nothing is hidden.
Reactive vs. autonomous — when to use which#
| Use reactive (SpaceChat) | Use autonomous (Agent / Workflow) | |
|---|---|---|
| Cadence | One-off | Recurring or event-driven |
| Need | Discussion, exploration | Execution, output |
| Autonomy | You stay in the loop | The system runs by itself |
| Output | Lives in the chat | Lives as entities in your tree |
A common rhythm: explore in SpaceChat, find a pattern worth keeping, promote it to an Agent.
Where to start#
- New to River AI? Open SpaceChat with Cmd J and start a conversation.
- Done something three times? Time for an Agent.
- Need a rigorous pipeline? Build a Workflow.
- Pattern shows up across many Agents? Extract a Skill.
Models#
River works with cloud models (Anthropic, OpenAI, xAI, Gemini, Venice) and local models via Ollama. The model picker is in the SpaceChat input bar; the Intelligence Router decides where each request actually runs based on capability, privacy, and cost.
When you're ready to bring AI fully on-device, see Local Models.