Tinode is an IM router and a store. Conceptually it loosely follows a [publish-subscribe](https://en.wikipedia.org/wiki/Publish%E2%80%93subscribe_pattern) model.
Server connects sessions, users, and topics. Session is a network connection between a client application and the server. User represents a human being who connects to the server with a session. Topic is a named communication channel which routes content between sessions.
Users and topics are assigned unique IDs. User ID is a string with 'usr' prefix followed by base64-URL-encoded pseudo-random 64-bit number, e.g. `usr2il9suCbuko`. Topic IDs are described below.
Clients such as mobile or web applications create sessions by connecting to the server over a websocket or through long polling. Client authentication is required in order to perform most operations. Client authenticates the session by sending a `{login}` packet. See [Authentication](#authentication) section for details. Once authenticated, the client receives a token which is used for authentication later. Multiple simultaneous sessions may be established by the same user. Logging out is not supported (and not needed).
* Peer to peer topic is a communication channel strictly between two users. Each participant sees topic name as the ID of the other participant: 'usr' prefix followed by a base64-URL-encoded numeric part of user ID, e.g. `usr2il9suCbuko`.
* Group topic is a channel for multi-user communication. It's named as 'grp' followed by 11 pseudo-random characters, i.e. `grpYiqEXb4QY6s`. Group topics must be explicitly created.
Session joins a topic by sending a `{sub}` packet. Packet `{sub}` serves three functions: creating a new topic, subscribing user to a topic, and attaching session to a topic. See [`{sub}`](#sub) section below for details.
Once the session has joined the topic, the user may start generating content by sending `{pub}` packets. The content is delivered to other attached sessions as `{data}` packets.
The user may query or update topic metadata by sending `{get}` and `{set}` packets.
Changes to topic metadata, such as changes in topic description, or when other users join or leave the topic, is reported to live sessions with `{pres}` (presence) packet. The `{pres}` packet is sent either to the topic being affected or to the `me` topic.
When user's `me` topic comes online (i.e. an authenticated session attaches to `me` topic), a `{pres}` packet is sent to `me` topics of all other users, who have peer to peer subscriptions with the first user.
Timestamps are always represented as [RFC 3339](http://tools.ietf.org/html/rfc3339)-formatted string with precision up to milliseconds and timezone always set to UTC, e.g. `"2015-10-06T18:07:29.841Z"`.
Whenever base64 encoding is mentioned, it means base64 URL encoding with padding characters stripped, see [RFC 4648](http://tools.ietf.org/html/rfc4648).
The `{data}` packets have server-issued sequential IDs: base-10 numbers starting at 1 and incrementing by one with every message. They are guaranteed to be unique per topic.
In order to connect requests to responses, client may assign message IDs to all packets set to the server. These IDs are strings defined by the client. Client should make them unique at least per session. The client-assigned IDs are not interpreted by the server, they are returned to the client as is.
There are three ways to access the server over the network: websocket, long polling, and [gRPC](https://grpc.io/).
When the client establishes a connection to the server over HTTP(S), such as over a websocket or long polling, the server offers the following endpoints:
Once the connection is opened, the client must issue a `{hi}` message to the server. Server responds with a `{ctrl}` message which indicates either success or an error. The `params` field of the response contains server's protocol version `"params":{"ver":"0.15"}` and may include other values.
See definition of the gRPC API in the [proto file](../pbx/model.proto). gRPC API has slightly more functionality than the API described in this document: it allows the `root` user to send messages on behalf of other users as well as delete users.
The `bytes` fields in protobuf messages expect JSON-encoded UTF-8 content. For example, a string should be quoted before being converted to bytes as UTF-8: `[]byte("\"some string\"")` (Go), `'"another string"'.encode('utf-8')` (Python 3).
Messages are sent in text frames, one message per frame. Binary frames are reserved for future use. By default server allows connections with any value in the `Origin` header.
Long polling works over `HTTP POST` (preferred) or `GET`. In response to client's very first request server sends a `{ctrl}` message containing `sid` (session ID) in `params`. Long polling client must include `sid` in every subsequent request either in the URL or in the request body.
Large files are sent out of band using `HTTP POST` as `Content-Type: multipart/form-data`. See [below](#out-of-band-handling-of-large-files) for details.
Tinode server can be set up to run behind a reverse proxy, such as NGINX. For efficiency it can accept client connections from Unix sockets by setting `listen` and/or `grpc_listen` config parameters to the path of the Unix socket file, e.g. `unix:/run/tinode.sock`. The server may also be configured to read peer's IP address from `X-Forwarded-For` HTTP header by setting `use_x_forwarded_for` config parameter to `true`.
Users are generally assigned one of the two authentication levels: authenticated `auth` or anonymous `anon`. The third level `root` is only accessible over `gRPC` where it permits the `root` to send messages on behalf of other users.
When a connection is first established, the client application can send either an `{acc}` or a `{login}` message which authenticates the user at one the levels.
Each user is assigned a unique ID. The IDs are composed as `usr` followed by base64-encoded 64-bit numeric value, e.g. `usr2il9suCbuko`. Users also have the following properties:
*`username`: unique string used in `basic` authentication; username is not accessible to other users
*`defacs`: object describing user's default access mode for peer to peer conversations with authenticated and anonymous users; see [Access control](#access-control) for details
*`auth`: default access mode for authenticated `auth` users
*`anon`: default access for anonymous `anon` users
User's account has a state. The following states are defined:
*`ok` (normal): the default state which means the account is not restricted in any way and can be used normally;
*`susp` (suspended): the user is prevented from accessing the account as well as not found through [search](#fnd-and-tags-finding-users-and-topics); the state can be assigned by the administrator and fully reversible.
*`del` (soft-deleted): user is marked as deleted but user's data is retained; un-deleting the user is not currenly supported.
*`undef` (undefined): used internally by authenticators; should not be used elsewhere.
A user may maintain multiple simultaneous connections (sessions) with the server. Each session is tagged with a client-provided `User Agent` string intended to differentiate client software.
Logging out is not supported by design. If an application needs to change the user, it should open a new connection and authenticate it with the new user credentials.
Authentication is conceptually similar to [SASL](https://en.wikipedia.org/wiki/Simple_Authentication_and_Security_Layer): it's provided as a set of adapters each implementing a different authentication method. Authenticators are used during account registration [`{acc}`](#acc) and during [`{login}`](#login). The server comes with the following authentication methods out of the box:
The `token` is intended to be the primary means of authentication. Tokens are designed in such a way that token authentication is light weight. For instance, token authenticator generally does not make any database calls, all processing is done in-memory. All other authentication methods are intended to be used only to obtain or refresh the token. Once the token is obtained, subsequent logins should use it.
The `basic` authentication scheme expects `secret` to be a base64-encoded string of a string composed of a user name followed by a colon `:` followed by a plan text password. User name in the `basic` scheme must not contain the colon character `:` (ASCII 0x3A).
The `anonymous` scheme can be used to create accounts, it cannot be used for logging in: a user creates an account using `anonymous` scheme and obtains a cryptographic token which it uses for subsequent `token` logins. If the token is lost or expired, the user is no longer able to access the account.
Compiled-in authenticator names may be changed by using `logical_names` configuration feature. For example, a custom `rest` authenticator may be exposed as `basic` instead of default one or `token` authenticator could be hidden from users. The feature is activated by providing an array of mappings in the config file: `logical_name:actual_name` to rename or `actual_name:` to hide. For instance, to use a `rest` service for basic authentication use `"logical_names": ["basic:rest"]`.
When a new account is created, the user must inform the server which authentication method will be later used to gain access to this account as well as provide shared secret, if appropriate. Only `basic` and `anonymous` can be used during account creation. The `basic` requires the user to generate and send a unique login and password to the server. The `anonymous` does not exchange secrets.
User may optionally set `{acc login=true}` to use the new account for immediate authentication. When `login=false` (or not set), the new account is created but the authentication status of the session which created the account remains unchanged. When `login=true` the server will attempt to authenticate the session with the new account, the `{ctrl}` response to the `{acc}` request will contain the authentication token on success. This is particularly important for the `anonymous` authentication because that's the only time when the authentication token can be retrieved.
Logging in is performed by issuing a `{login}` request. Logging in is possible with `basic` and `token` only. Response to any login is a `{ctrl}` message with either a code 200 and a token which can be used in subsequent logins with `token` authentication, or a code 300 request for additional information, such as verifying credentials or responding to a method-dependent challenge in multi-step authentication, or a code 4xx error.
User may change authentication parameters, such as changing login and password, by issuing an `{acc}` request. Only `basic` authentication currently supports changing parameters:
If the session is not authenticated, the request must include a `token`. It can be a regular authentication token obtained during login, or a restricted token received through [Resetting a Password](#resetting-a-password) process. If the session is authenticated, the token must not be included. If the request is authenticated for access level `ROOT`, then the `user` may be set to a valid ID of another user. Otherwise it must be blank (defaulting to the current user) or equal to the ID of the current user.
To reset login or password, (or any other authentication secret, if such action is supported by the authenticator), one sends a `{login}` message with the `scheme` set to `reset` and the `secret` containing a base64-encoded string "`authentication scheme to reset secret for`:`reset method`:`reset method value`". Most basic case of resetting a password by email is
If the email matches the registration, the server will send a message using specified method and address with instructions for resetting the secret. The email contains a restricted security token which the user can include into an `{acc}` request with the new secret as described in [Changing Authentication Parameters](#changing-authentication-parameters).
Sending the same message with `status: "ok"` un-suspends the account. A root user may check account status by executing `{get what="desc"}` command against user's `me` topic.
Server may be optionally configured to require validation of certain credentials associated with the user accounts and authentication scheme. For instance, it's possible to require user to provide a unique email or a phone number, or to solve a captcha as a condition of account registration.
The server supports verification of email out of the box with just a configuration change. is mostly functional, verification of phone numbers is not functional because a commercial subscription is needed in order to be able to send text messages (SMS).
If certain credentials are required, then user must maintain them in validated state at all times. It means if a required credential has to be changed, the user must first add and validate the new credential and only then remove the old one.
Credentials are initially assigned at registration time by sending an `{acc}` message, added using `{set topic="me"}`, deleted using `{del topic="me"}`, and queries by `{get topic="me"}` messages. Credentials are verified by the client by sending either a `{login}` or an `{acc}` message.
Access control manages user's access to topics through access control lists (ACLs). The access is assigned individually to each user-topic pair (subscription).
Access control is mostly usable for group topics. Its usability for `me` and P2P topics is limited to managing presence notifications and banning uses from initiating or continuing P2P conversations. All channel readers are given the same permissions.
User's access to a topic is defined by two sets of permissions: user's desired permissions "want", and permissions granted to user by topic's manager(s) "given". Each permission is represented by a bit in a bitmap. It can be either present or absent. The actual access is determined as a bitwise AND of wanted and given permissions. The permissions are communicated in messages as a set of ASCII characters, where presence of a character means a set permission bit:
* No access: `N` is not a permission per se but an indicator that permissions are explicitly cleared/not set. It usually indicates that the default permissions should *not* be applied.
* Join: `J`, permission to subscribe to a topic
* Read: `R`, permission to receive `{data}` packets
* Write: `W`, permission to `{pub}` to topic
* Presence: `P`, permission to receive presence updates `{pres}`
* Owner: `O`, user is the topic owner; the owner can assign any other permission to any topic member, change topic description, delete topic; topic may have a single owner only; some topics have no owner
When a user subscribes to a topic or starts a chat with another user, the access permissions are either set explicitly or assigned by default `defacs`. Access permissions can be modified by sending `{set}` messages.
A client may set explicit permissions in `{sub}` and `{set}` messages. If the permissions are missing or set to an empty string (not `N`!), Tinode will use default permissions `defacs` assigned earlir. If no default permissions are found, the authenticated users in group topics will receive a `JRWPS` access, in P2P topics will get `JRWPA`; anonymous users will receive `N` (no access) which means every subscription request must be explicitly approved by the topic manager.
Default access is defined for two categories of users: authenticated and anonymous. The default access value is applied as a "given" permission to all new subscriptions. Topic's default access is established at the topic creation time by `{sub.desc.defacs}` and can be subsequently modified by the owner by sending `{set}` messages. Likewise, user's default access is established at the account creation time by `{acc.desc.defacs}` and can be modified by the user by sending a `{set}` message to `me` topic.
Topic is a named communication channel for one or more people. Topics have persistent properties. These topic properties can be queried by `{get what="desc"}` message.
Topic usually have subscribers. One of the subscribers may be designated as topic owner (`O` access permission) with full access permissions. The list of subscribers can be queries with a `{get what="sub"}` message. The list of subscribers is returned in a `sub` section of a `{meta}` message.
Topic `me` is automatically created for every user at the account creation time. It serves as means of managing account information, receiving presence notification from people and topics of interest. Topic `me` has no owner. The topic cannot be deleted or unsubscribed from. One can `leave` the topic which will stop all relevant communication and indicate that the user is offline (although the user may still be logged in and may continue to use other topics).
Joining or leaving `me` generates a `{pres}` presence update sent to all users who have peer to peer topics with the given user and `P` permissions set.
Message `{get what="desc"}` to `me` is automatically replied with a `{meta}` message containing `desc` section with the topic parameters (see intro to [Topics](#topics) section). The `public` parameters of `me` topic is data that the user wants to show to his/her connections. Changing it changes `public` not just for the `me` topic, but also everywhere where user's `public` is shown, such as `public` of all user's peer to peer topics.
Message `{get what="sub"}` to `me` is different from any other topic as it returns the list of topics that the current user is subscribed to as opposite to the expected user's subscription to `me`.
* seq: server-issued numeric id of the last message in the topic
Topic `fnd` is automatically created for every user at the account creation time. It serves as an endpoint for discovering other users and group topics. Users and group topics can be discovered by `tags`. Tags are optionally assigned at the topic or user creation time then can be updated by using `{set what="tags"}` against a `me` or a group topic.
A tag is an arbitrary case-insensitive Unicode string (forced to lowercase on the server) up to 96 characters long which may contain characters from `Letter` and `Number` Unicode [classes/categories](https://en.wikipedia.org/wiki/Unicode_character_property#General_Category) as well as any of the following ASCII characters: `_`, `.`, `+`, `-`, `@`, `#`, `!`, `?`.
Tag may have a prefix which serves as a namespace. The prefix is a 2-16 character string which starts with a letter [a-z] and may contain lowercase ASCII letters and numbers followed by a colon `:`, ex. prefixed phone tag `tel:+14155551212` or prefixed email tag `email:alice@example.com`. Some prefixed tags are optionally enforced to be unique. In that case only one user or topic may have such a tag. Certain tags may be forced to be immutable to the user, i.e. user's attempts to add or remove an immutable tag will be rejected by the server.
The tags are indexed server-side and used in user and topic discovery. Search returns users and topics sorted by the number of matched tags in descending order.
In order to find users or topics, a user sets either `public` or `private` parameter of the `fnd` topic to a search query (see [Query language](#query-language)) then issues a `{get topic="fnd" what="sub"}` request. If both `public` and `private` are set, the `public` query is used. The `private` query is persisted across sessions and devices, i.e. all user's sessions see the same `private` query. The value of the `public` query is ephemeral, i.e. it's not saved to database and not shared between user's sessions. The `private` query is intended for large queries which do not change often, such as finding matches for everyone in user's contact list on a mobile phone. The `public` query is intended to be short and specific, such as finding some topic or a user who is not in the contact list.
_CURRENTLY UNSUPPORTED_ When a new user registers with tags matching the given query, the `fnd` topic will receive `{pres}` notification for the new user.
Tinode query language is used to define search queries for finding users and topics. The query is a string containing atomic terms separated by spaces or commas. The individual query terms are matched against user's or topic's tags. The individual terms may be written in an RTL language but the query as a whole is parsed left to right. Spaces are treated as the `AND` operator, commas (as well as commas preceded and/or followed by a space) as the `OR` operator. The order of operators is ignored: all `AND` tags are grouped together, all `OR` tags are grouped together. `OR` takes precedence over `AND`: if a tag is preceded of followed by a comma, it's an `OR` tag, otherwise an `AND`. For example, `aaa bbb, ccc` (`aaa AND bbb OR ccc`) is interpreted as `(bbb OR ccc) AND aaa`.
*`flowers, travel puppies, kittens`: find topics or users which contain either one of `flowers`, `travel`, `puppies`, or `kittens`, i.e. `flowers OR travel OR puppies OR kittens`. The space between `travel` and `puppies` is treated as `OR` due to `OR` taking precedence over `AND`.
_CURRENTLY UNSUPPORTED_ Queries, particularly `fnd.private` could be arbitrarily large, limited only by the limits on the message size, and by the limits on the query size in the underlying database. Instead of rewriting the entire query to add or remove a term, terms can be added or removed incrementally.
Finding users by login, phone or email requires query terms to be written with prefixes, i.e. `email:alice@example.com` instead of `alice@example.com`. This may present a problem to end users because it requires them to learn the query language. Tinode solves this problem by implementing _query rewrite_ on the server: if query term (tag) does not contain a prefix, server rewrites it by adding the appropriate prefix. In queries to `fnd.public` the original term is also kept (query `alice@example.com` is rewritten as `email:alice@example.com OR alice@example.com`), in queries to `fnd.private` only the rewritten term is kept (`alice@example.com` is rewritten as `email:alice@example.com`). All terms that look like email, for instance, `alice@example.com` are rewritten to `email:alice@example.com OR alice@example.com`. Terms which look like phone numbers are converted to [E.164](https://en.wikipedia.org/wiki/E.164) and also rewritten as `tel:+14155551212 OR +14155551212`. In addition, in queries to `fnd.public` all other unprefixed terms which look like logins are rewritten as logins: `alice` -> `basic:alice OR alice`.
As described above, tags which look like phone numbers are converted to E.164 format. Such conversion requires an ISO 3166-1 alpha-2 country code. The following logic is used when converting phone number tags to E.164:
* If the tag already contains a country calling code, it's used as is: `+1(415)555-1212` -> `+14155551212`.
An immutable tag(s) may be assigned to the user which denotes the organisation the user belongs to. When the user searches for other users or topics, the search can be restricted to always contain the tag. This approach can be used to segment users into organisations with limited visibility into each other.
Client software may periodically assign a [geohash](https://en.wikipedia.org/wiki/Geohash) tag to the user based on current location. Searching for users in a given area would mean matching on geohash tags.
The approach is similar to geohashing. The entire range of numbers is covered by the smallest possible power of 2, for instance the range of human ages is covered by 2<sup>7</sup>=128 years. The entire range is split in two halves: the range 0-63 is denoted by 0, 64-127 by 1. The operation is repeated with each subrange, i.e. 0-31 is 00, 32-63 is 01, 0-15 is 000, 32-47 is 010. Once completed, the age 30 will belong to the following ranges: 0 (0-63), 00 (0-31), 001 (16-31), 0011 (24-31), 00111 (28-31), 001111 (30-31), 0011110 (30). A 30 y.o. user is assigned a few tags to indicate the age, i.e. `age:00111`, `age:001111`, and `age:0011110`. Technically, all 7 tags may be assigned but usually it's impractical. To query for anyone in the age range 28-35 convert the range into a minimal number of tags: `age:00111` (28-31), `age:01000` (32-35). This query will match the 30 y.o. user by tag `age:00111`.
Peer to peer (P2P) topics represent communication channels between strictly two users. The name of the topic is different for each of the two participants. Each of them sees the name of the topic as the user ID of the other participant: `usr` followed by base64 URL-encoded ID of the user. For example, if two users `usrOj0B3-gSBSs` and `usrIU_LOVwRNsc` start a P2P topic, the first one will see it as `usrIU_LOVwRNsc`, the second as `usrOj0B3-gSBSs`. The P2P topic has no owner.
A P2P topic is created by one user subscribing to topic with the name equal to the ID of the other user. For instance, user `usrOj0B3-gSBSs` can establish a P2P topic with user `usrIU_LOVwRNsc` by sending a `{sub topic="usrIU_LOVwRNsc"}`. Tinode will respond with a `{ctrl}` packet with the name of the newly created topic as described above. The other user will receive a `{pres}` message on `me` topic with updated access permissions.
The 'public' parameter of P2P topics is user-dependent. For instance a P2P topic between users A and B would show user A's 'public' to user B and vice versa. If a user updates 'public', all user's P2P topics will automatically update 'public' too.
Group topic represents a communication channel between multiple users. The name of a group topic is `grp` or `chn` followed by a string of characters from base64 URL-encoding set. No other assumptions can be made about internal structure or length of the group name.
Group topics support limited number of subscribers (controlled by a `max_subscriber_count` parameter in configuration file) with access permissions of each subscriber managed individually. Group topics may also be enabled to support any number of read-only users - `readers`. All `readers` have the same access permissions. Group topics with enabled `readers` are called `channels`.
A group topic is created by sending a `{sub}` message with the topic field set to string `new` or `nch` optionally followed by any characters, e.g. `new` or `newAbC123` are equivalent. Tinode will respond with a `{ctrl}` message with the name of the newly created topic, i.e. `{sub topic="new"}` is replied with `{ctrl topic="grpmiKBkQVXnm3P"}`. If topic creation fails, the error is reported on the original topic name, i.e. `new` or `newAbC123`. The user who created the topic becomes topic owner. Ownership can be transferred to another user with a `{set}` message but one user must remain the owner at all times.
A `channel` topic is different from the non-channel group topic in the following ways:
* Channel topic is created by sending `{sub topic="nch"}`. Sending `{sub topic="new"}` will create a group topic without enabling channel functionality.
* Sending `{sub topic="chnAbC123"}` will create a `reader` subscription to a channel. A non-channel topic will reject such subscription request.
* When searching for topics using [`fnd`](#fnd-and-tags-finding-users-and-topics), channels will show addresses with `chn` prefixes, non-channel topic will show with `grp` prefixes.
* A subscriber joining or leaving the topic (regular or channel-enabled) generates a `{pres}` message to all other subscribers who are currently in the joined state with the topic and have appropriate permissions. Reader joining or leaving the channel generates no `{pres}` message.
The `sys` topic serves as an always available channel of communication with the system administrators. A normal non-root user cannot subscribe to `sys` but can publish to it without subscription. Existing clients use this channel to report abuse by sending a Drafty-formatted `{pub}` message with the report as JSON attachment. A root user can subscribe to `sys` topic. Once subscribed, the root user will receive messages sent to `sys` topic by other users.
Tinode provides basic support for client-side caching of `{data}` messages in the form of server-issued sequential message IDs. The client may request the last message id from the topic by issuing a `{get what="desc"}` message. If the returned ID is greater than the ID of the latest received message, the client knows that the topic has unread messages and their count. The client may fetch these messages using `{get what="data"}` message. The client may also paginate history retrieval by using message IDs.
A user is reported as being online when one or more of user's sessions are attached to the `me` topic. Client-side software identifies itself to the server using `ua` (user agent) field of the `{login}` message. The _user agent_ is published in `{meta}` and `{pres}` messages in the following way:
* When user's first session attaches to `me`, the _user agent_ from that session is broadcast in the `{pres what="on" ua="..."}` message.
* When multiple user sessions are attached to `me`, the _user agent_ of the session where the most recent action has happened is reported in `{pres what="ua" ua="..."}`; the 'action' in this context means any message sent by the client. To avoid potentially excessive traffic, user agent changes are broadcast no more frequently than once a minute.
* When user's last session detaches from `me`, the _user agent_ from that session is recorded together with the timestamp; the user agent is broadcast in the `{pres what="off" ua="..."}` message and subsequently reported as the last online timestamp and user agent.
An empty `ua=""`_user agent_ is not reported. I.e. if user attaches to `me` with non-empty _user agent_ then does so with an empty one, the change is not reported. An empty _user agent_ may be disallowed in the future.
Topics and subscriptions have `trusted`, `public`, and `private` fields. Generally, the fields are application-defined. The server does not enforce any particular structure of these fields except for `fnd` topic. At the same time, client software should use the same format for interoperability reasons. The following sections describe the format of these fields as they are implemented by all official clients.
The format of the optional `trusted` field in group and peer to peer topics is a set of key-value pairs; `fnd` and `sys` topics do not have the `trusted`. The following optional keys are currently defined:
Although it's not yet enforced, custom fields should start with an `x-` followed by the application name, e.g. `x-myapp-value: "abc"`. The fields should contain primitive types only, i.e. `string`, `boolean`, `number`, or `null`.
Format of `content` field in `{pub}` and `{data}` is application-defined and as such the server does not enforce any particular structure of the field. At the same time, client software should use the same format for interoperability reasons. Currently the following two types of `content` are supported:
Tinode provides two endpoints for handling large files: `/v0/file/u` for uploading files and `v0/file/s` for downloading. The endpoints require the client to provide both [API key](#connecting-to-the-server) and login credentials. The server checks credentials in the following order:
To upload a file first create an RFC 2388 multipart request then send it to the server using HTTP POST. The server responds to the request either with a `307 Temporary Redirect` with the new upload URL, or with a `200 OK` and a `{ctrl}` message in response body:
If `307 Temporary Redirect` is returned, the client must retry the upload at the provided URL. The URL returned in `307` response should be used for just this one upload. All subsequent uploads should try the default URL first.
The `ctrl.params.url` contains the path to the uploaded file at the current server. It could be either the full path like `/v0/file/s/mfHLxDWFhfU.pdf`, a relative path like `./mfHLxDWFhfU.pdf`, or just the file name `mfHLxDWFhfU.pdf`. Anything but the full path is interpreted against the default *download* endpoint `/v0/file/s/`. For instance, if `mfHLxDWFhfU.pdf` is returned then the file is located at `http(s)://current-tinode-server/v0/file/s/mfHLxDWFhfU.pdf`.
Once the URL of the file is received, either immediately or after following the redirect, the client may use the URL to send a `{pub}` message with the uploaded file as an attachment, or, if the file is an image, as an avatar image for a topic or user profile (see [theCard](./thecard.md)). For example, the URL can be used in a [Drafty](./drafty.md)-formatted `pub.content` field:
It's important to list the used URLs in the `extra: attachments[...]` field. Tinode server uses this field to maintain the uploaded file's use counter. Once the counter drops to zero for the given file (for instance, because a message with the shared URL was deleted or because the client failed to include the URL in the `extra.attachments` field), the server will garbage collect the file. Only relative URLs should be used. Absolute URLs in the `extra.attachments` field are ignored. The URL value is expected to be the `ctrl.params.url` returned in response to upload.
The serving endpoint `/v0/file/s` serves files in response to HTTP GET requests. The client must evaluate relative URLs against this endpoint, i.e. if it receives a URL `mfHLxDWFhfU.pdf` or `./mfHLxDWFhfU.pdf` it should interpret it as a path `/v0/file/s/mfHLxDWFhfU.pdf` at the current Tinode HTTP server.
_Important!_ As a security measure, the client should not send security credentials if the download URL is absolute and leads to another server.
Tinode uses compile-time adapters for handling push notifications. The server comes with [Tinode Push Gateway](../server/push/tnpg/), [Google FCM](https://firebase.google.com/docs/cloud-messaging/), and `stdout` adapters. Tinode Push Gateway and Google FCM support Android with [Play Services](https://developers.google.com/android/guides/overview) (may not be supported by some Chinese phones), iOS devices and all major web browsers excluding Safari. The `stdout` adapter does not actually send push notifications. It's mostly useful for debugging, testing and logging. Other types of push notifications such as [TPNS](https://intl.cloud.tencent.com/product/tpns) can be handled by writing appropriate adapters.
Tinode Push Gateway (TNPG) is a proprietary Tinode service which sends push notifications on behalf of Tinode. Internally it uses Google FCM and as such supports the same platforms as FCM. The main advantage of using TNPG over FCM is simplicity of configuration: mobile clients do not need to be recompiled, all is needed is a [configuration update](../server/push/tnpg/) on a server.
[Google FCM](https://firebase.google.com/docs/cloud-messaging/) supports Android with [Play Services](https://developers.google.com/android/guides/overview), iPhone and iPad devices, and all major web browsers excluding Safari. In order to use FCM mobile clients (iOS, Android) must be recompiled with credentials obtained from Google. See [instructions](../server/push/fcm/) for details.
The `stdout` adapter is mostly useful for debugging and logging. It writes push payload to `STDOUT` where it can be redirected to file or read by some other process.
Tinode provides an optional service which helps client applications generate link (URL) previews for inclusion into messages. The enpoint of this service (if enabled) is located at `/v0/urlpreview`. The service takes a single parameter `url`:
```
/v0/urlpreview?url=https%3A%2F%2Ftinode.co
```
The first several kilobytes of the document at the given URL is fetched by issuing an HTTP(S) GET request. If the returned document has content-type `text/html`, the HTML is parsed for page title, description, and image URL. The result is formatted as JSON and returned as
```json
{"title": "Page title", "description": "This is a page description", "image_url": "https://tinode.co/img/logo64x64.png"}
```
The link preview service requires authentication. It's exactly the same as authentication for [Out of Band Large Files](#out-of-band-handling-of-large-files).
All client to server messages may have an optional `id` field. It's set by the client as means to receive an acknowledgement from the server that the message was received and processed. The `id` is expected to be a session-unique string but it can be any string. The server does not attempt to interpret it other than to check JSON validity. The `id` is returned unchanged by the server when it replies to the client message.
Server requires strictly valid JSON, including double quotes around field names. For brevity the notation below omits double quotes around field names as well as outer curly brackets. Examples use `//` comments only for expressiveness. The comments cannot be used in actual communication with the server.
data needs to be cleared, use a string with a single Unicode DEL character "␡" (`\u2421`). I.e. sending `"public": null` will not clear the field, but sending `"public": "␡"` will.
Every client to server message contains the main payload described in the sections below and an optional top-level field `extra`:
```js
{
abc: { ... }, // Main payload, see sections below.
extra: {
attachments: ["/v0/file/s/sJOD_tZDPz0.jpg"], // Array of out-of-band attachments which have to be exempted from GC.
obo: "usr2il9suCbuko", // Alternative user ID set by the root user (obo = On Behalf Of).
authlevel: "auth" // Altered authentication level set by the root user.
}
}
```
The `attachments` array lists URLs of files uploaded out of band. Such listing increments use counter of these files. Once the use counter drops to 0, the files will be automatically deleted.
The `obo` (On Behalf Of) can be set by the `root` user. If the `obo` is set, the server will treat the message as if it came from the specified user as opposite to the actual sender.
The `authlevel` is supplementary to the `obo` and permits setting custom authentication level for the user. A an `"auth"` level is used if the field is unset.
The user agent `ua` is expected to follow [RFC 7231 section 5.5.3](http://tools.ietf.org/html/rfc7231#section-5.5.3) recommendation but the format is not enforced. The message can be sent more than once to update `ua`, `dev` and `lang` values. If sent more than once, the `ver` field of the second and subsequent messages must be either unchanged or not set.
Message `{acc}` creates users or updates `tags` or authentication credentials `scheme` and `secret` of exiting users. To create a new user set `user` to the string `new` optionally followed by any character sequence, e.g. `newr15gsr`. Either authenticated or anonymous session can send an `{acc}` message to create a new user. To update authentication data or validate a credential of the current user leave `user` unset.
Server responds with a `{ctrl}` message with `params` containing details of the new user account such as user ID and, in case of `login: true`, authentication token. If `desc.defacs` is missing, the server will assign server-default access permissions to new account.
Server responds to a `{login}` packet with a `{ctrl}` message. The `params` of the message contains the id of the logged in user as `user`. The `token` contains an encrypted string which can be used for authentication. Expiration time of the token is passed as `expires`.
User creates a new group topic by sending `{sub}` packet with the `topic` field set to `new12321` (regular topic) or `nch12321` (channel) where `12321` denotes any string including an empty string. Server will create a topic and respond back to the session with the name of the newly created topic.
If the user had no relationship with the topic, sending `{sub}` packet creates it. Subscribing means to establish a relationship between session's user and the topic where no relationship existed in the past.
Joining (attaching to) a topic means for the session to start consuming content from the topic. Server automatically differentiates between subscribing and joining/attaching based on context: if the user had no prior relationship with the topic, the server subscribes the user then attaches the current session to the topic. If relationship existed, the server only attaches the session to the topic. When subscribing, the server checks user's access permissions against topic's access control list. It may grant immediate access, deny access, may generate a request for approval from topic managers.
The `{sub}` message may include a `get` and `set` fields which mirror `{get}` and `{set}` messages. If included, server will treat them as a subsequent `{set}` and `{get}` messages on the same topic. If the `get` is set, the reply may include `{meta}` and `{data}` messages.
Server responds to `{leave}` with a `{ctrl}` packet. Leaving without unsubscribing affects just the current session. Leaving with unsubscribing will affect all user's sessions.
Topic subscribers receive the `content` in the [`{data}`](#data) message. By default the originating session gets a copy of `{data}` like any other session currently attached to the topic. If for some reason the originating session does not want to receive the copy of the data it just published, set `noecho` to `true`.
*`replace`: an indicator that the message is a correction/replacement for another message, a topic-unique ID of the message being updated/replaced, `":123"`
*`sender`: a user ID of the sender added by the server when the message is sent on behalf of another user, `"usr1XUtEhjv6HND"`.
*`thread`: an indicator that the message is a part of a conversation thread, a topic-unique ID of the first message in the thread, `":123"`; `thread` is intended for tagging a flat list of messages as opposite to creating a tree.
Application-specific fields should start with an `x-<application-name>-`. Although the server does not enforce this rule yet, it may start doing so in the future.
The unique message ID should be formed as `<topic_name>:<seqId>` whenever possible, such as `"grp1XUtEhjv6HND:123"`. If the topic is omitted, i.e. `":123"`, it's assumed to be the current topic.
Query topic for metadata, such as description or a list of subscribers, or query message history. The requester must be [subscribed and attached](#sub) to the topic to receive the full response. Some limited `desc` and `sub` information is available without being attached.
Query message history. Server sends `{data}` messages matching parameters provided in the `data` field of the query.
The `id` field of the data messages is not provided as it's common for data messages. When all `{data}` messages are transmitted, a `{ctrl}` message is sent.
Query [credentials](#credentail-validation). Server responds with a `{meta}` message containing an array of credentials. Supported for `me` topic only.
Update topic metadata, delete messages or topic. The requester is generally expected to be [subscribed and attached](#sub) to the topic. Only `desc.private` and requester's `sub.mode` can be updated without attaching first.
User can soft-delete `hard=false` (default) or hard-delete `hard=true` messages. Soft-deleting messages hides them from the requesting user but does not delete them from storage. An `R` permission is required to soft-delete messages. Hard-deleting messages deletes message content from storage (`head`, `content`) leaving a message stub. It affects all users. A `D` permission is needed to hard-delete messages. Messages can be deleted in bulk by specifying one or more message ID ranges in `delseq` parameter. Each delete operation is assigned a unique `delete ID`. The greatest `delete ID` is reported back in the `clear` of the `{meta}` message.
`what="sub"`
Deleting a subscription removes specified user from topic subscribers. It requires an `A` permission. A user cannot delete own subscription. A `{leave}` should be used instead. If the subscription is soft-deleted (default), it's marked as deleted without actually deleting a record from storage.
`what="topic"`
Deleting a topic deletes the topic including all subscriptions, and all messages. Only the owner can delete a topic.
`what="user"`
Deleting a user is a very heavy operation. Use caution.
Delete credential. Validated credentials and those with no attempts at validation are hard-deleted. Credentials with failed attempts at validation are soft-deleted which prevents their reuse by the same user.
Client-generated ephemeral notification for forwarding to other clients currently attached to the topic, such as typing notifications or delivery receipts. The message is "fire and forget": not stored to disk per se and not acknowledged by the server. Messages deemed invalid are silently dropped.
The `{note.recv}` and `{note.read}` do alter persistent state on the server. The value is stored and reported back in the corresponding fields of the `{meta.sub}` message.
The `read` and `recv` notifications may optionally include `unread` value which is the total count of unread messages as determined by this client. The per-user `unread` count is maintained by the server: it's incremented when new `{data}` messages are sent to user and reset to the values reported by the `{note unread=...}` message. The `unread` value is never decremented by the server. The value is included in push notifications to be shown on a badge on iOS:
Data messages have a `seq` field which holds a sequential numeric ID generated by the server. The IDs are guaranteed to be unique within a topic. IDs start from 1 and sequentially increment with every successful [`{pub}`](#pub) message received by the topic.
Tinode uses `{pres}` message to inform clients of important events. A separate [document](https://docs.google.com/spreadsheets/d/e/2PACX-1vStUDHb7DPrD8tF5eANLu4YIjRkqta8KOhLvcj2precsjqR40eDHvJnnuuS3bw-NcWsP1QKc7GSTYuX/pubhtml?gid=1959642482&single=true) explains all possible use cases.
The `{pres}` messages are purely transient: they are not stored and no attempt is made to deliver them later if the destination is temporarily unavailable.
Forwarded client-generated notification `{note}`. Server guarantees that the message complies with this specification and that content of `topic` and `from` fields is correct. The other content is copied from the `{note}` message verbatim and may potentially be incorrect or misleading if the originator so desires.
