Parisneo Lollms Vs Coder Lollms Client Discussion Mastering

1. Data Zones: The Context Layers

LoLLMs uses a layered context approach. When you call get_full_data_zone(), it assembles these layers in specific order:

1. Memory: Long-term facts.
2. User Data Zone: Global user preferences (e.g., "I prefer Python 3.10").
3. Discussion Data Zone: Metadata for the specific task (e.g., "Currently working on Chapter 3").
4. Personality Data Zone: Temporary data from tools (e.g., RAG search results).
5. Active Artefacts: The content of versioned documents.

# Direct manipulation of session context
discussion.user_data_zone = "User Role: Senior Architect"
discussion.discussion_data_zone = "Project Goal: Build a FastAPI backend."

2. The Memory System

Memory can be manually set or automatically extracted using memorize().

• Manual: Best for explicit facts.
• Auto-Memorize: Asks the LLM to extract technical content and solutions from the history.

# Auto-extract facts from current conversation
memory_entry = discussion.memorize() 
# Returns {"title": "...", "content": "..."} if successful

3. Versioned Artefact System

Artefacts are persistent documents that survive conversation turns. They support metadata and full version control.

A. Creation & Metadata

Always use XML tags. The system stores attributes like author, source, and description.

<artefact name="outline.md" type="document" author="ParisNeo" description="Book Structure">
# Book Outline
1. Introduction
2. The Awakening
3. The Conflict
</artefact>

B. Efficient Patching (Aider Format)

To save context, use Aider blocks. The backend finds the SEARCH text and replaces it with REPLACE.

<artefact name="outline.md">
<<<<<<< SEARCH
2. The Awakening
=======
2. The New Awakening (Expanded)
>>>>>>> REPLACE
</artefact>

C. Versioning & Reverting

Every update increments the version. You can see history counts in the context and revert via tags:

<!-- Revert 'outline.md' to version 1 -->
<revert_artefact name="outline.md" version="1" />

4. Agentic Chat & UI Notifications

When using discussion.chat(), the backend emits a MSG_TYPE_ARTEFACTS_STATE_CHANGED notification whenever an artefact is created, updated, or reverted.

from lollms_client import MSG_TYPE

def my_callback(text, msg_type, meta):
    if msg_type == MSG_TYPE.MSG_TYPE_ARTEFACTS_STATE_CHANGED:
        # meta["artefacts"] contains the updated list of dictionaries
        print(f"Artefact updated: {text}")

discussion.chat(
    user_message="Update the second chapter of the book.",
    streaming_callback=my_callback
)

5. Practical Workflow: Writing a Book

1. Initialization: Ask LLM to create an artefact with an empty structure.
2. Iterative Filling: Ask LLM to fill specific segments using aider patches.
3. Refinement: Use <revert_artefact> if a change wasn't desired.
4. Context Management: Use discussion.summarize_and_prune(max_tokens=4096) regularly to keep the history lean while the Artefact keeps the primary content stable.

Source: ParisNeo/lollms-vs-coder — distributed by TomeVault.