Markers

Markers are recognized by Bitrix24 AI module and interpreted for the neural network in use.

Basic markers

{original_message} – original message for preprompt operation. Text editor architecture contains the complete typed text or a portion of typed text, highlighted manually.

Preprompt example:

Continue text: {orignal_message}

The result: text is continued based on the text typed above or a manually highlighted portion of text.

{user_message} – the text directly inputted by user.

Preprompt example:

Write a story about {user_message}

These markers can also be applied jointly:

Write a story about {user_message} to continue the text: {original_message}.

{role}{/role} – binary marker specifying neural network and its role. Inside you can enter a text in a free format that gives the network some instructions. For example, "speak as robot". On one hand it seems that you can simply write the same in a preprompt. On the other hand, instructions can have a certain structure, albeit customizable and neural networks receive them better.

The 'role' marker specifics of use:

• You can use other markers inside the input instructions;
• Only the first instruction is used, the rest of them are removed;
• Marker is added after if-conditions, meaning you can use a single block {role} for 'if' and a second block for 'else'.

Example:

Tell a joke. {role}Chat as a robot named {author.name}{/role}

Additional markers

{author_message} – text for the original Feed post or task description.

Example:

Highlight the synapsis based on text: {author_message}

{context_messages} – "fit as much as possible" context. For example, an original post + all the recent comments or chat history for a specific period. Or task description + comments.

Example:

Generate a task result based on its description and comments posted to it: {context_messages}

All the basic markers may be expanded by developers in the future commercial releases.

{language} – portal language (for the user language). For example, English, German, Spanish – marker will be replaced specifically to theses values.

Preprompt example:

Translate text to {language}. Text: {original_message}

Author's markers

When we work with context-related messages, each such message has an author with a work position and other characteristics. Because an author is also a user, you can use the user sting fields. You can view the list of such fields at documentation page for CUser class.

When writing a marker you need to consider a one rule for formatting. All positions must be in lower case: {author.lower_case_field}.

Example. Use the fields NAME and WORK_PISITION in preprompt for the author:

Give a compliment to the author.
Post you need to compliment: {original_message}
Post author's name: {author.name}
Post author's work position: {author.work_position}

Markers for current result

A mini-conversation occurs when CoPilot is in operation. You are asking a question (or edit the text) and receive a result with which you can continue to working with.

This produces a set of results that you can query in markers. Currently, stack size is three messages.

This set has several specific details:

1. Results are supplemented from top to bottom, indicating the most recent result from the bottom up;
2. The most reсent result is a current one inside the response at the moment of dispatched new query (and not the future result to be received);
3. Numbering starts from zero.

Example

{current_result0} – result, visible at the moment of sending a query (employed prompt);
{current_result1} – the same previous result;
{current_result2} – the result before the previous one.

Note. If some markers are still empty, they are removed from text.