Ravi for organizations

Run agents on as many machines as your teams need, and connect them all to one Console organization. Each machine keeps its own agents, credentials and history. The organization sees the agents each operator lists, lets agents on different machines ask each other, and shares one Library of skills, agents, themes and widgets. Each machine runs the Ravi daemon and sets its own permission ceilings. The Ravi Console holds the organization: members, projects, the Library, and the list of machines and their agents. What leaves your machines lists everything a machine sends, and when.

This page describes ravi.bot@next.

How it fits together

A machine joins with ravi login and works in one organization, the one picked in the browser. From then on its daemon calls the Console on a schedule. It reports the agents its operator listed, reads the machine’s Library pins and picks up requests from other machines. No agent is listed by default.

What lives where

Agents, conversations, and model and channel credentials stay on the machine. The Console keeps what people and machines share. Tokens of connected accounts, such as Gmail, are the exception: they stay in Ravi’s hosted connector service.

On each machine	In the Console organization
The daemon and the channel runner, under PM2	Members, their roles and pending invites
Agents: instructions (`AGENTS.md`), folder, provider and model	Projects, with their Pages and Bases
Sessions, message history and traces, in SQLite under `~/.ravi`	The Library: skills, agent definitions, themes and widgets
Instances, routes, contacts, triggers and settings (`ravi.db`)	The machines linked with `ravi login`
Model credentials and channel tokens	The agents listed on each machine: id, name, provider, model, project, description and whether it takes requests
Permission ceilings, denials and audit events	Which Library skill is pinned to which agent
Cost and token metrics	Requests between machines, page shares and access requests

The agent report never carries instructions, paths or credentials. Session traces, with each turn’s tokens and cost, stay on the machine by default. They are uploaded only when you set RAVI_SYNC_RUNNER_ENABLED=1 or run ravi sync push --traces --execute.

What works today and what does not

Linking machines, sharing a Library and asking across machines work now. Machine identity owned by the organization does not exist yet.

Capability	Status	How it works today
Machines join the organization	Ready	`ravi login` sends the machine’s name, hostname, platform, Ravi version and a random key kept on the machine.
The Console knows each machine’s agents	Ready	An agent shows only after its operator lists it with `ravi agents listing <agent> on --execute`. While logged in with the `console.projects.link` scope, the daemon reports up to 500 listed agents at start and every 5 minutes. `ravi agents directory` lists them, online or offline.
Agents ask agents on other machines	Ready	`ravi agents ask <machine>/<agent> "<message>" --execute`. The asked agent answers as the person who asked, not as its operator.
Opening an agent to requests	Ready	Every agent is closed by default. Its operator opens it with `ravi agents requests <agent> on --execute`, which also lists it.
Organization and project Library	Ready	Skills pinned to an agent load on its machine; a new pin needs the agent listed. Agent definitions install with `ravi agents install`. Themes and widgets apply when a page ships.
Members and access	Ready	Five roles (`owner`, `admin`, `developer`, `viewer`, `billing`), invites, sharing a private page by email, and access requests.
Members-only data pages	Ready	Live Bases data reaches only signed-in members. `ravi pages ship` refuses a public route that reads Bases, unless you pass `--members-best-effort`.
Governance on each machine	Ready	A permission ceiling per agent, the write brake (exit 3 until `--execute`), and denials recorded in `ravi.db` and published on `ravi.audit.denied`.
Machine identity owned by the organization	Not yet	Today a machine is linked by the Console user who ran `ravi login`. Admin approval and revocation that survives a new login are not built.
Fleet view	Not yet	The Console has no health or metrics view of your machines. `ravi agents directory` shows each agent as online or offline.
Pages chat with the organization’s agents	Not yet	Page Chat lets a signed-in member talk by voice to an agent that operates the page. It does not reach the agents on your machines.

Open source and hosted

The daemon, the CLI and the SDKs are MIT-licensed and run without the Console. The organization, Pages, Bases, the Library and requests between machines need console.ravi.bot.

Part	License	Needs the Console
Daemon, CLI and channel runner	MIT	No
SDKs for TypeScript, Dart and Swift, and the HTTP gateway	MIT	No
Sessions, cron, triggers, heartbeats, permissions and local audit	MIT, part of the daemon	No
The organization, members, invites, page sharing and access requests	Hosted	Yes
Pages, Bases, the Library, the agent directory and requests between machines	Hosted	Yes

Machines: link a machine, name it and list its agents.
Data flow: what each machine sends to the Console, and when.
Security: ceilings, the write brake, audit and credentials.
Log in to the Console: create the organization and run ravi login.

Start here

Connect a channel

Console and Pages

For organizations

Build on Ravi

Ravi for organizations

How it fits together

What lives where

What works today and what does not

Open source and hosted

Next

​How it fits together

​What lives where

​What works today and what does not

​Open source and hosted

​Next

How it fits together

What lives where

What works today and what does not

Open source and hosted

Next