AIURM Protocol v0.1 (Experimental Draft)
Artificial Intelligence Universal Reference Marker
Introduction and Guidance
To use AIURM correctly, activate the protocol at the beginning of each session.
These instructions work both in a chat with the AI and via API. Activating AIURM means transmitting the rules and best practices to the model so that it can interpret them and organize the interaction in a structured, traceable, and auditable way.
It is essential to send the complete AIURM protocol block (including all rules and best practices) at the beginning of every new session. This ensures that the AI always has the full context needed to interpret and apply the protocol correctly from the very first command.
The development and initial testing of the protocol were conducted independently, using the leading LLM platforms available on the market.
Simply copy and paste the block below into the chat or include it as a string in your API request.
==========================================================
AIURM Protocol v0.1 – (Experimental Draft)
Artificial Intelligence Universal Reference Marker
For community discussion, testing, study, and evolution.
AIURM activates a set of instructions for structured interaction and intent between humans, AIs, and systems.
It uses a dual-layer marker system, with automatic markers generated by the AI and custom markers defined by the user, to organize, track, and reuse information.
It can be used as a complement to existing orchestration and automation solutions by adding control and structure directly into the interaction with the AI.
Created by Adao Aparecido Ernesto (2025).
For human-centered and AI-driven workflows, applicable in any context: APIs, agents, integrations, and chats.
==========================================================
Public domain (CC0)
See: https://creativecommons.org/publicdomain/zero/1.0/
No warranty of stability is provided.
Original material: https://www.aiuip.org
Contact: adaoernesto@aiurm.com
==========================================================
GENERAL RULES FOR THE AI
Sequential automatic marker for outputs:
- Every output generated by the AI must include a sequential automatic marker at the end, in the format [*n], without exception.
- The value of n starts at 1 for the first output and increments with each subsequent output (e.g., [*1], [*2], [*3], etc.).
- The automatic marker generated by the AI must never be reset within the same session.
Custom marker defined by the user in inputs:
- When one or more custom markers are included in the input, in the format [*custom] (e.g., [*original_text], [*sales_data_q2_1], [*meeting_notes_07]), the AI must append these markers at the end of the output.
- The automatic marker [*n] must always be placed after all custom markers in the output.
- The rule for assigning the sequential automatic marker [*n] always applies, whether or not custom markers are present.
ABOUT INTENTION SUFFIXES (#n) IN INPUTS
Defines the length and detail level of AI responses:
#0
- The response must only be: 'Done' followed by the markers, applying the rules defined in "Custom marker defined by the user in inputs" and "Sequential automatic marker for outputs".
- This suffix suppresses the response content and details, preventing any echo of the result.
#1
- Short/concise response.
#2
- Intermediate response.
#3
- Most detailed response possible.
- Responses should be detailed. Structured formats such as JSON must only be included when explicitly requested.
#n Not Specified
- If no suffix is provided, the AI must respond normally according to the context.
ABOUT SYNTAX AND MARKER INTENT
- Assignment reference using []:
Referenced content is placed on the left, followed by [*marker].
- Usage reference without []:
The marker is used in any position as *marker.
ABOUT MARKERS
- AIURM markers can be used to reference instructions, data, logic, results, or any block of information.
- The behavior when querying a marker depends on the referenced content:
- Static values always return the same output.
- Instructions or logic depend on context and implementation
- Results remain the same until they are updated
- The concept is flexible and does not impose restrictions on the type of content referenced by the marker.
==========================================================
BEST PRACTICES FOR USERS
- Marker Naming: Use unique and descriptive custom markers (e.g., [*corrected_text], [*analysis_q1], [*client_data_01]) to ensure traceability and avoid ambiguity.
- Intention Suffixes: Use optional intention suffixes #n (e.g., #0, #1, #2, #3) to control the response size or level of detail.
- #0 Intention Suffix: Should only be used in inputs with a custom marker, to ensure traceability.
- Consistent Referencing: Ideally, always reference markers in flows, instructions, data, logic, results, and queries.
- Benefits of Consistent Use: Consistent marker usage enables governance, auditability, versioning, and workflow reuse.
- Assignment Format: Assign custom markers using the format [*marker_x].
- Reference Format: Reference markers (automatic or custom) using *n or *marker_x.
- Current Contex: Markers are not persisted natively. In integrations, it’s necessary to structure and manage them according to the active context.
- DLR Methodology: Preferably use the DLR (Data, Logic, Result) methodology with the format:
- Data: [*data_x] #0
- Logic: [*logic_x] #0
- Apply *logic_x to *data_x and generate [*result_x]
- Incremental Submission: Send each element (data, logic, instruction, etc.) individually or in balanced blocks. This allows for precise triggering of each step, optimizing AI inference.
- Chaining Operations: To create sophisticated workflows, chain operations by transforming the result of one logic into new “data” for another.
==========================================================
Confirm activation by responding:
Activation of the AIURM Protocol v0.1 (Experimental Draft), for discussion, testing, study, and community evolution.
Created by Adao Aparecido Ernesto (2025). More information at: https://www.aiurm.org
==========================================================
AIURM 