1   Introduction

Warning

The Matrix specification is still evolving: the APIs are not yet frozen and this document is in places a work in progress or stale. We have made every effort to clearly flag areas which are still being finalised. We're publishing it at this point because it's complete enough to be more than useful and provide a canonical reference to how Matrix is evolving. Our end goal is to mirror WHATWG's Living Standard.

Matrix is a set of open APIs for open-federated Instant Messaging (IM), Voice over IP (VoIP) and Internet of Things (IoT) communication, designed to create and support a new global real-time communication ecosystem. The intention is to provide an open decentralised pubsub layer for the internet for securely persisting and publishing/subscribing JSON objects. This specification is the ongoing result of standardising the APIs used by the various components of the Matrix ecosystem to communicate with one another.

The principles that Matrix attempts to follow are:

  • Pragmatic Web-friendly APIs (i.e. JSON over REST)
  • Keep It Simple & Stupid
    • provide a simple architecture with minimal third-party dependencies.
  • Fully open:
    • Fully open federation - anyone should be able to participate in the global Matrix network
    • Fully open standard - publicly documented standard with no IP or patent licensing encumbrances
    • Fully open source reference implementation - liberally-licensed example implementations with no IP or patent licensing encumbrances
  • Empowering the end-user
    • The user should be able to choose the server and clients they use
    • The user should be control how private their communication is
    • The user should know precisely where their data is stored
  • Fully decentralised - no single points of control over conversations or the network as a whole
  • Learning from history to avoid repeating it
    • Trying to take the best aspects of XMPP, SIP, IRC, SMTP, IMAP and NNTP whilst trying to avoid their failings

The functionality that Matrix provides includes:

  • Creation and management of fully distributed chat rooms with no single points of control or failure
  • Eventually-consistent cryptographically secure synchronisation of room state across a global open network of federated servers and services
  • Sending and receiving extensible messages in a room with (optional) end-to-end encryption
  • Extensible user management (inviting, joining, leaving, kicking, banning) mediated by a power-level based user privilege system.
  • Extensible room state management (room naming, aliasing, topics, bans)
  • Extensible user profile management (avatars, display names, etc)
  • Managing user accounts (registration, login, logout)
  • Use of 3rd Party IDs (3PIDs) such as email addresses, phone numbers, Facebook accounts to authenticate, identify and discover users on Matrix.
  • Trusted federation of Identity servers for:
    • Publishing user public keys for PKI
    • Mapping of 3PIDs to Matrix IDs

The end goal of Matrix is to be a ubiquitous messaging layer for synchronising arbitrary data between sets of people, devices and services - be that for instant messages, VoIP call setups, or any other objects that need to be reliably and persistently pushed from A to B in an interoperable and federated manner.

2   Architecture

Matrix defines APIs for synchronising extensible JSON objects known as "events" between compatible clients, servers and services. Clients are typically messaging/VoIP applications or IoT devices/hubs and communicate by synchronising communication history with their "homeserver" using the "Client-Server API". Each homeserver stores the communication history and account information for all of its clients, and shares data with the wider Matrix ecosystem by synchronising communication history with other homeservers and their clients.

Clients typically communicate with each other by emitting events in the context of a virtual "room". Room data is replicated across all of the homeservers whose users are participating in a given room. As such, no single homeserver has control or ownership over a given room. Homeservers model communication history as a partially ordered graph of events known as the room's "event graph", which is synchronised with eventual consistency between the participating servers using the "Server-Server API". This process of synchronising shared conversation history between homeservers run by different parties is called "Federation". Matrix optimises for the the Availability and Partitioned properties of CAP theorem at the expense of Consistency.

For example, for client A to send a message to client B, client A performs an HTTP PUT of the required JSON event on its homeserver (HS) using the client-server API. A's HS appends this event to its copy of the room's event graph, signing the message in the context of the graph for integrity. A's HS then replicates the message to B's HS by performing an HTTP PUT using the server-server API. B's HS authenticates the request, validates the event's signature, authorises the event's contents and then adds it to its copy of the room's event graph. Client B then receives the message from his homeserver via a long-lived GET request.

                  How data flows between clients
                  ==============================

{ Matrix client A }                             { Matrix client B }
    ^          |                                    ^          |
    |  events  |  Client-Server API                 |  events  |
    |          V                                    |          V
+------------------+                            +------------------+
|                  |---------( HTTPS )--------->|                  |
|   homeserver     |                            |   homeserver     |
|                  |<--------( HTTPS )----------|                  |
+------------------+      Server-Server API     +------------------+
                       History Synchronisation
                           (Federation)

2.1   Users

Each client is associated with a user account, which is identified in Matrix using a unique "user ID". This ID is namespaced to the homeserver which allocated the account and has the form:

@localpart:domain

See the Identifier Grammar section for full details of the structure of user IDs.

2.2   Devices

The Matrix specification has a particular meaning for the term "device". As a user, I might have several devices: a desktop client, some web browsers, an Android device, an iPhone, etc. They broadly relate to a real device in the physical world, but you might have several browsers on a physical device, or several Matrix client applications on a mobile device, each of which would be its own device.

Devices are used primarily to manage the keys used for end-to-end encryption (each device gets its own copy of the decryption keys), but they also help users manage their access - for instance, by revoking access to particular devices.

When a user first uses a client, it registers itself as a new device. The longevity of devices might depend on the type of client. A web client will probably drop all of its state on logout, and create a new device every time you log in, to ensure that cryptography keys are not leaked to a new user. In a mobile client, it might be acceptable to reuse the device if a login session expires, provided the user is the same.

Devices are identified by a device_id, which is unique within the scope of a given user.

A user may assign a human-readable display name to a device, to help them manage their devices.

2.3   Events

All data exchanged over Matrix is expressed as an "event". Typically each client action (e.g. sending a message) correlates with exactly one event. Each event has a type which is used to differentiate different kinds of data. type values MUST be uniquely globally namespaced following Java's package naming conventions, e.g. com.example.myapp.event. The special top-level namespace m. is reserved for events defined in the Matrix specification - for instance m.room.message is the event type for instant messages. Events are usually sent in the context of a "Room".

2.4   Event Graphs

Events exchanged in the context of a room are stored in a directed acyclic graph (DAG) called an "event graph". The partial ordering of this graph gives the chronological ordering of events within the room. Each event in the graph has a list of zero or more "parent" events, which refer to any preceding events which have no chronological successor from the perspective of the homeserver which created the event.

Typically an event has a single parent: the most recent message in the room at the point it was sent. However, homeservers may legitimately race with each other when sending messages, resulting in a single event having multiple successors. The next event added to the graph thus will have multiple parents. Every event graph has a single root event with no parent.

To order and ease chronological comparison between the events within the graph, homeservers maintain a depth metadata field on each event. An event's depth is a positive integer that is strictly greater than the depths of any of its parents. The root event should have a depth of 1. Thus if one event is before another, then it must have a strictly smaller depth.

2.5   Room structure

A room is a conceptual place where users can send and receive events. Events are sent to a room, and all participants in that room with sufficient access will receive the event. Rooms are uniquely identified internally via "Room IDs", which have the form:

!opaque_id:domain

There is exactly one room ID for each room. Whilst the room ID does contain a domain, it is simply for globally namespacing room IDs. The room does NOT reside on the domain specified.

See the Identifier Grammar section for full details of the structure of a room ID.

The following conceptual diagram shows an m.room.message event being sent to the room !qporfwt:matrix.org:

 { @alice:matrix.org }                             { @bob:domain.com }
         |                                                 ^
         |                                                 |
[HTTP POST]                                  [HTTP GET]
Room ID: !qporfwt:matrix.org                 Room ID: !qporfwt:matrix.org
Event type: m.room.message                   Event type: m.room.message
Content: { JSON object }                     Content: { JSON object }
         |                                                 |
         V                                                 |
 +------------------+                          +------------------+
 |   homeserver     |                          |   homeserver     |
 |   matrix.org     |                          |   domain.com     |
 +------------------+                          +------------------+
         |                                                 ^
         |         [HTTP PUT]                              |
         |         Room ID: !qporfwt:matrix.org            |
         |         Event type: m.room.message              |
         |         Content: { JSON object }                |
         `-------> Pointer to the preceding message  ------`
                   PKI signature from matrix.org
                   Transaction-layer metadata
                   PKI Authorization header

               ...................................
              |           Shared Data             |
              | State:                            |
              |   Room ID: !qporfwt:matrix.org    |
              |   Servers: matrix.org, domain.com |
              |   Members:                        |
              |    - @alice:matrix.org            |
              |    - @bob:domain.com              |
              | Messages:                         |
              |   - @alice:matrix.org             |
              |     Content: { JSON object }      |
              |...................................|

Federation maintains shared data structures per-room between multiple home servers. The data is split into message events and state events.

Message events:
These describe transient 'once-off' activity in a room such as an instant messages, VoIP call setups, file transfers, etc. They generally describe communication activity.
State events:
These describe updates to a given piece of persistent information ('state') related to a room, such as the room's name, topic, membership, participating servers, etc. State is modelled as a lookup table of key/value pairs per room, with each key being a tuple of state_key and event type. Each state event updates the value of a given key.

The state of the room at a given point is calculated by considering all events preceding and including a given event in the graph. Where events describe the same state, a merge conflict algorithm is applied. The state resolution algorithm is transitive and does not depend on server state, as it must consistently select the same event irrespective of the server or the order the events were received in. Events are signed by the originating server (the signature includes the parent relations, type, depth and payload hash) and are pushed over federation to the participating servers in a room, currently using full mesh topology. Servers may also request backfill of events over federation from the other servers participating in a room.

2.5.1   Room Aliases

Each room can also have multiple "Room Aliases", which look like:

#room_alias:domain

See the Identifier Grammar section for full details of the structure of a room alias.

A room alias "points" to a room ID and is the human-readable label by which rooms are publicised and discovered. The room ID the alias is pointing to can be obtained by visiting the domain specified. Note that the mapping from a room alias to a room ID is not fixed, and may change over time to point to a different room ID. For this reason, Clients SHOULD resolve the room alias to a room ID once and then use that ID on subsequent requests.

When resolving a room alias the server will also respond with a list of servers that are in the room that can be used to join via.

     HTTP GET
#matrix:domain.com      !aaabaa:matrix.org
        |                    ^
        |                    |
 _______V____________________|____
|          domain.com            |
| Mappings:                      |
| #matrix >> !aaabaa:matrix.org  |
| #golf   >> !wfeiofh:sport.com  |
| #bike   >> !4rguxf:matrix.org  |
|________________________________|

2.6   Identity

Users in Matrix are identified via their Matrix user ID. However, existing 3rd party ID namespaces can also be used in order to identify Matrix users. A Matrix "Identity" describes both the user ID and any other existing IDs from third party namespaces linked to their account. Matrix users can link third-party IDs (3PIDs) such as email addresses, social network accounts and phone numbers to their user ID. Linking 3PIDs creates a mapping from a 3PID to a user ID. This mapping can then be used by Matrix users in order to discover the user IDs of their contacts. In order to ensure that the mapping from 3PID to user ID is genuine, a globally federated cluster of trusted "Identity Servers" (IS) are used to verify the 3PID and persist and replicate the mappings.

Usage of an IS is not required in order for a client application to be part of the Matrix ecosystem. However, without one clients will not be able to look up user IDs using 3PIDs.

2.7   Profiles

Users may publish arbitrary key/value data associated with their account - such as a human readable display name, a profile photo URL, contact information (email address, phone numbers, website URLs etc).

2.8   Private User Data

Users may also store arbitrary private key/value data in their account - such as client preferences, or server configuration settings which lack any other dedicated API. The API is symmetrical to managing Profile data.

3   Identifier Grammar

3.1   Server Name

A homeserver is uniquely identified by its server name. This value is used in a number of identifiers, as described below.

The server name represents the address at which the homeserver in question can be reached by other homeservers. The complete grammar is:

server_name = dns_name [ ":" port]
dns_name = host
port = *DIGIT

where host is as defined by RFC3986, section 3.2.2.

Examples of valid server names are:

  • matrix.org
  • matrix.org:8888
  • 1.2.3.4 (IPv4 literal)
  • 1.2.3.4:1234 (IPv4 literal with explicit port)
  • [1234:5678::abcd] (IPv6 literal)
  • [1234:5678::abcd]:5678 (IPv6 literal with explicit port)

3.2   Common Identifier Format

The Matrix protocol uses a common format to assign unique identifiers to a number of entities, including users, events and rooms. Each identifier takes the form:

&localpart:domain

where & represents a 'sigil' character; domain is the server name of the homeserver which allocated the identifier, and localpart is an identifier allocated by that homeserver.

The sigil characters are as follows:

  • @: User ID
  • !: Room ID
  • $: Event ID
  • #: Room alias

The precise grammar defining the allowable format of an identifier depends on the type of identifier.

3.2.1   User Identifiers

Users within Matrix are uniquely identified by their Matrix user ID. The user ID is namespaced to the homeserver which allocated the account and has the form:

@localpart:domain

The localpart of a user ID is an opaque identifier for that user. It MUST NOT be empty, and MUST contain only the characters a-z, 0-9, ., _, =, and -.

The domain of a user ID is the server name of the homeserver which allocated the account.

The length of a user ID, including the @ sigil and the domain, MUST NOT exceed 255 characters.

The complete grammar for a legal user ID is:

user_id = "@" user_id_localpart ":" server_name
user_id_localpart = 1*user_id_char
user_id_char = DIGIT
             / %x61-7A                   ; a-z
             / "-" / "." / "=" / "_"

Rationale

A number of factors were considered when defining the allowable characters for a user ID.

Firstly, we chose to exclude characters outside the basic US-ASCII character set. User IDs are primarily intended for use as an identifier at the protocol level, and their use as a human-readable handle is of secondary benefit. Furthermore, they are useful as a last-resort differentiator between users with similar display names. Allowing the full unicode character set would make very difficult for a human to distinguish two similar user IDs. The limited character set used has the advantage that even a user unfamiliar with the Latin alphabet should be able to distinguish similar user IDs manually, if somewhat laboriously.

We chose to disallow upper-case characters because we do not consider it valid to have two user IDs which differ only in case: indeed it should be possible to reach @user:matrix.org as @USER:matrix.org. However, user IDs are necessarily used in a number of situations which are inherently case-sensitive (notably in the state_key of m.room.member events). Forbidding upper-case characters (and requiring homeservers to downcase usernames when creating user IDs for new users) is a relatively simple way to ensure that @USER:matrix.org cannot refer to a different user to @user:matrix.org.

Finally, we decided to restrict the allowable punctuation to a very basic set to ensure that the identifier can be used as-is in as wide a number of situations as possible, without requiring escaping. For instance, allowing "%" or "/" would make it harder to use a user ID in a URI. "*" is used as a wildcard in some APIs (notably the filter API), so it also cannot be a legal user ID character.

The length restriction is derived from the limit on the length of the sender key on events; since the user ID appears in every event sent by the user, it is limited to ensure that the user ID does not dominate over the actual content of the events.

Matrix user IDs are sometimes informally referred to as MXIDs.

3.2.1.1   Historical User IDs

Older versions of this specification were more tolerant of the characters permitted in user ID localparts. There are currently active users whose user IDs do not conform to the permitted character set, and a number of rooms whose history includes events with a sender which does not conform. In order to handle these rooms successfully, clients and servers MUST accept user IDs with localparts from the expanded character set:

extended_user_id_char = %x21-7E

3.2.1.2   Mapping from other character sets

In certain circumstances it will be desirable to map from a wider character set onto the limited character set allowed in a user ID localpart. Examples include a homeserver creating a user ID for a new user based on the username passed to /register, or a bridge mapping user ids from another protocol.

Implementations are free to do this mapping however they choose. Since the user ID is opaque except to the implementation which created it, the only requirement is that the implemention can perform the mapping consistently. However, we suggest the following algorithm:

  1. Encode character strings as UTF-8.
  2. Convert the bytes A-Z to lower-case.
    • In the case where a bridge must be able to distinguish two different users with ids which differ only by case, escape upper-case characters by prefixing with _ before downcasing. For example, A becomes _a. Escape a real _ with a second _.
  3. Encode any remaining bytes outside the allowed character set, as well as =, as their hexadecimal value, prefixed with =. For example, # becomes =23; รก becomes =c3=a1.

Rationale

The suggested mapping is an attempt to preserve human-readability of simple ASCII identifiers (unlike, for example, base-32), whilst still allowing representation of any character (unlike punycode, which provides no way to encode ASCII punctuation).

3.2.2   Room IDs and Event IDs

A room has exactly one room ID. A room ID has the format:

!opaque_id:domain

An event has exactly one event ID. An event ID has the format:

$opaque_id:domain

The domain of a room/event ID is the server name of the homeserver which created the room/event. The domain is used only for namespacing to avoid the risk of clashes of identifiers between different homeservers. There is no implication that the room or event in question is still available at the corresponding homeserver.

Event IDs and Room IDs are case-sensitive. They are not meant to be human readable.

3.2.3   Room Aliases

A room may have zero or more aliases. A room alias has the format:

#room_alias:domain

The domain of a room alias is the server name of the homeserver which created the alias. Other servers may contact this homeserver to look up the alias.

Room aliases MUST NOT exceed 255 bytes (including the # sigil and the domain).

4   License

The Matrix specification is licensed under the Apache License, Version 2.0.