Agents

An Agent is a reusable AI specialist that lives in your workspace. A Deal Researcher. An Inbox Triage Analyst. A Morning Briefer. A Weekly Investor Update Drafter. Each one has a name, a job, a set of tools it can use, and a way to wake up.

Agents are what make River feel like it's working for you, not just with you. They notice, act, and report — whether or not you have a tab open.

Anatomy of an Agent#

Every Agent has four things:

• A name and persona — its identity and systemPrompt (how it should think and write)
• A toolset — the specific tools it's allowed to use (read mail, write docs, post to Slack…)
• One or more triggers — when it should run (event, schedule, manual)
• A home — it lives in your entity tree, just like any doc

You can open an Agent in a Window, edit it, share it with a teammate, fork it, or delete it. It's a first-class entity.

How Agents run#

When an Agent is triggered:

1. River starts a Job to track the run.
2. The Agent loads its persona, the relevant context, and its allowed tools.
3. It runs the agentic loop: think → call a tool → see the result → think again.
4. It writes results back as entities — drafts, docs, tasks, replies — and posts to its chat thread.
5. The Job completes (or fails) and the run is logged in Activity.

You can watch a Job in real time, cancel one in progress, and replay any past run.

Three kinds of triggers#

Agents wake up in three ways. See Triggers for full detail.

• World-event triggers — react when something happens (an email arrives, a file lands, a task completes, an integration connects).
• Schedule triggers — run on a cadence (every weekday at 7am, every Monday morning).
• Manual triggers — run on demand from the AutomationsApp, the API, or by clicking "Run now."

A single Agent can have multiple triggers. Your Morning Briefer might run on a 7am schedule and be invokable manually.

Agents vs. SpaceChat#

Use SpaceChat for ad-hoc conversation. Promote a recurring task to an Agent the second it stops being a one-off.

Where Agents live#

Open the AutomationsApp (or press Cmd K → "Automations") to see all Agents in your Dataspace. From there you can:

• Create a new Agent
• Edit any Agent's persona, tools, or triggers
• Run an Agent manually
• See the Job history of every run
• Pause or delete Agents
• Share an Agent with the rest of your team

Examples worth stealing#

• Inbox Triage Analyst — wakes on every email.received, classifies the message, drafts a reply for high-priority threads, files the rest.
• Deal Researcher — wakes when an email arrives from a known investor; pulls fund data, checks shared connections, drafts a one-pager.
• Morning Briefer — runs at 7am on weekdays; assembles yesterday's progress, today's calendar, and three priorities into a doc.
• Standup Synthesizer — runs after every team standup transcript lands; turns it into action items and updates the project doc.
• Customer Support First-Pass — wakes on every new ticket; drafts a reply with citations and routes the rest.

Next#

→ Build your first one in Building an Agent. → Pick the right trigger for the job in Triggers. → Decide between an Agent and a Workflow in the Automation Mental Model.