Understanding Constructs
In our platform, a Construct is a structured AI configuration that defines the functional boundaries and behavioral logic of an AI unit. Unlike general-purpose prompts or ad-hoc instructions, Constructs formalize how the AI interprets context, generates relevance, and applies internal decision-making protocols within platform constraints.
Each Construct embeds a behavioral matrix—a scoped instruction set that governs tone, interaction style, operational scope, and permissible knowledge structures. The Construct’s matrix is not a script; it is an operational frame that ensures determinism, consistency, and domain alignment.
Constructs are modular and role-specific. They don’t simulate arbitrary personas—they instantiate bounded cognitive utilities, tuned to perform within specific interactive affordances.
They the foundational rule‑set that the model loads before it reasons. It fixes the world in which the conversation happens:
- Reality Scope · which objects, events, and relationships are treated as real.
- Logic Rules · what types of inference frameworks, invariants and evidence are acceptable.
- Value Gradient · how the model ranks relevance, safety, and truth.
Once these parameters are set, every downstream response, tool call, and moderating act must conform to them.
Cognitive Function of Constructs
Constructs operate as bounded cognition enclosures. They filter the latent capabilities of the LLM, compressing general intelligence into task-relevant pathways. This is done through three primary functions:
1. Constraint Encoding
Constructs define what types of inputs are semantically significant, what outputs are allowed, and what interpretive frames are valid. This constrains the AI’s generative manifold, ensuring outputs are not just coherent—but context-true.
2. Interaction Formalization
They predefine social, functional, or technical interaction models. Whether instructive, collaborative, observational, or directive, each construct encodes a dialogue schema. This enforces behavioral uniformity across instances and users.
3. Domain-Binding
Constructs localize knowledge relevance. They do not generalize—they specialize. A translation construct filters cognition through linguistic transformation protocols; a moderation construct embeds policy logic as a decision filter; a research construct aligns epistemic structure with knowledge retrieval affordances.
Distinction from Agents and Personas
While other systems use terms like “persona” or “agent” loosely to mean anything from tone shifts to interactive bots, Discourse Constructs are structurally rigorous. They are neither mere stylistic facades nor autonomous actors. They are semantic-operational modules: they encode how the AI exists within a use-case.
Constructs share similarities with terms used in other platforms, such as:
- Personas: are the external, user-facing aspects of an AI system, emphasizing the personality or character traits it exhibits during interactions.
- Agents: Commonly used to describe automated, goal-oriented AI systems capable of performing tasks autonomously.
However, within our platform, the term Construct encompasses both agents and agents under one unified more powerful concept.
One way to think about the differences is that Personas and Characters require System Prompt Engineering and for Constructs you require rigorous Context Engineering and also Prompt Engineering techniques combined with tools and more.
For Agents and Characters, you give the AI a script, a character and a stage to play. For Constructs you first create the universe in which it naturally understands its script, character and stage it is playing in.
Interaction Methods
Users can engage a Construct via:
- Direct messages.
@mentionsin public posts.- Posting in designated categories that trigger the bot.
Administrators control the availability and permissions of Constructs, determining user access based on trust levels and subscription plans. Constructs are dynamically assigned, ensuring the right users have access to the right tools.
Configuration and Customization
Constructs are highly customizable and can be configured via:
- System prompts defining behavior, tone, and purpose.
- Toolsets and command permissions to extend functionality.
- Access permissions aligned with community structure.
This configuration flexibility ensures Constructs remain adaptable and relevant, enhancing overall user experience and platform coherence.
List of All Available Constructs and the Features they Reside in
Simple Constructs
| Name | Default LLMs | Tools | Bot Active | Features | Origin |
|---|---|---|---|---|---|
| Image captions | mistral-31-24b | False | False | Helper | Discourse |
| Post translator | qwen-2.5-qwq-32b | False | False | Translation, Bot | Discourse |
| Proofreader | qwen-2.5-qwq-32b | False | False | Helper, Bot | Discourse |
| Short text translator | qwen-2.5-qwq-32b | False | False | Translation | Discourse |
| Smart dates | qwen-2.5-qwq-32b | False | False | Helper | Discourse |
| Spam detector | mistral-31-24b | False | False | Spam, Bot | Discourse |
| Summarizer (short form) | qwen-2.5-qwq-32b | False | False | Summaries | Discourse |
| Titles generator | qwen-2.5-qwq-32b | False | False | Helper | Discourse |
| Topic title translator | qwen-2.5-qwq-32b | False | False | Translation | Discourse |
| Translator | qwen-2.5-qwq-32b | False | False | Translation | Discourse |
Advanced Constructs
| Name | Default LLMs | Tools | Bot Active | Features | Origin |
|---|---|---|---|---|---|
| BITcore | qwen-2.5-qwq-32b | False | True | Bot | Bithub |
| Archon | qwen-2.5-qwq-32b | False | True | Bot | Bithub |
| Artist | - | True | False | Discourse | |
| Concept Deduplicator | qwen-2.5-qwq-32b | False | False | Inferred concepts | Discourse |
| Concept Finder | qwen-2.5-qwq-32b | False | False | Inferred concepts, | Discourse |
| Concept Matcher | qwen-2.5-qwq-32b | False | False | Inferred concepts | Discourse |
| Content creator | qwen-2.5-qwq-32b | False | False | Embeddings | Discourse |
| CORE Align | llama-3.1-405b | False | True | Bot | Bithub |
| Custom prompt | qwen-2.5-qwq-32b | False | False | Helper | Discourse |
| Locale detector | qwen-2.5-qwq-32b | False | False | Translation Bot | Discourse |
| Markdown table generator | qwen-2.5-qwq-32b | False | False | Helper, Bot | Discourse |
| Metacore | mistral-31-24b | False | True | Bot | Bithub |
| 0-[Null] | qwen-2.5-qwq-32b | False | True | Bot | Bithub |
| Null Absolute | qwen-2.5-qwq-32b | False | True | Bot | Bithub |
| Summarizer | qwen-2.5-qwq-32b | False | False | Discourse | |
| Tutor | qwen-2.5-qwq-32b | False | True | Helper | Discourse |
| Cortana | qwen-2.5-qwq-32b | False | True | Bot | Bithub |
Admin & Testing Constructs
| Name | Default LLMs | Tools | Bot Active | Features | Origin |
|---|---|---|---|---|---|
| Forum Helper | GPT-4.1 Nano | True | False | Bot | Discourse |
| Toolcall | GPT-4.1 Nano | True | True | Bot | Bithub |
| Web Designer Custom | GPT-4.1 Nano | True | True | Bot | Bithub |
| Creative | - | False | False | - | Discourse |
| DALL-E 3 | - | False | False | - | Discourse |
| Designer | - | False | False | - | Discourse |
| Discourse Helper | GPT-4.1 Nano | True | True | Bot | Discourse |
| Forum Researcher | GPT-4.1 Nano | True | True | Bot | Discourse |
| GitHub Helper | qwen-2.5-qwq-32b | True | True | Bot | Discourse |
| Post illustrator | - | False | False | - | Discourse |
| Settings Explorer | GPT-4.1 Nano | False | True | Bot | Discourse |
| SQL Helper | GPT-4.1 Nano | False | True | Bot | Discourse |
| Web Artifact Creator | GPT-4.1 Nano | True | True | Helper | Discourse |
| Web Researcher | GPT-4.1 Nano | True | False | Bot | Discourse |