Models, fine-tuning and LLM backends¶

What was said about this, newest first. For what exists today, see feature status.

Decisions, plans and what was built¶

12 Oct 2026: Said to exist · System terminal instances are controlled and monitored via MCP (Model Context Protocol) integration using Cmax.
12 Oct 2026: Said to exist · Decision making and multi-model judging are executed through OpenRouter using embedded rule engines to prevent greed, hatred, and delusion.
12 Oct 2026: Said to exist · The backend execution engine runs a Recursive Self-Improvement (RSI) loop optimized for agent creation and iterative fine-tuning.
5 Oct 2026: Said to exist · Speaker 1 uses cmux on macOS as an MCP-driven terminal tool to orchestrate Claude CLI sessions and agent swarms.
28 Sep 2026: Planned · The team should create an initial UI artifact prototype using Claude Design and share the artifact link with Speaker 1.
28 Sep 2026: Planned · The initial target workflow models a triad of teacher, student, and parent roles.
28 Sep 2026: Planned · The harness will be released as open-source before releasing the commercial Memory as a Service offering.
28 Sep 2026: Said to exist · Machine World's premium offering is a memory management service that prevents users from having to repeat instructions to models.
21 Sep 2026: Said to exist · Speaker 2 built an automated video generation pipeline using Claude to handle screen recording, automatic captions, and intro/outro voice stitching.
21 Sep 2026: Said to exist · An MCP server modeled on the [doctrine term] causal doctrine has been developed to handle memory decay and context cessation.
6 Jul 2026: Planned · Speaker 2 needs to create an agent to automate the setup of trial accounts and synthetic datasets for live testbed testing.
6 Jul 2026: Said to exist · Speaker 2 built the CLI interface using Somnet/Sonnet files without using Claude Design initially.
2 Mar 2026: Planned · Speaker 2 must capture the inputs, actual outputs, and expected outputs whenever the system fails and forward them to [name withheld] for fine-tuning.
23 Feb 2026: Planned · Speaker 1 will share resources and links regarding Claude/agent skills and MCP with [name withheld].
23 Feb 2026: Planned · Expose the validated website data via an MCP tool interface so external voice models and LLMs can accurately query and respond on Abhidhamma topics.
2 Feb 2026: Planned · [names withheld] will collaborate to build an automated reel generation agent using templates, style guides, and video model tokens.
19 Jan 2026: Planned · Transition to self-hosting custom models in later stages to eliminate variable third-party model costs.
29 Dec 2025: Said to exist · Speaker 2 uses Gboard voice typing to digitize Sinhala Abhidhamma texts because OCR solutions (including local LLaMA models) failed.
29 Dec 2025: Planned · Connect with [name withheld] and monastic teachers at Maggavihara to model [doctrine term] conditional relations between mental states in the next version.
8 Dec 2025: Planned · [name withheld] will generate English and Sinhala Dhamma analyses and musical descriptions for approximately 220 to 230 songs using Claude prompts.
22 Sep 2025: Said to exist · Speaker 1 regenerated and uploaded two older tracks using an updated AI model.

Ideas, questions and details (34)

5 Oct 2026: Decided · Speaker 2 tested background poster generation using the Hugging Face API and Flux model.
28 Sep 2026: Idea · The harness should load on top of OpenRouter to connect flexibly to any required underlying model or API.
28 Sep 2026: Decided · The system architecture is designed such that unwholesome mental states like greed, hatred, and delusion cannot naturally manifest in model behavior.
21 Sep 2026: Open · Local Sinhala AI language modeling research remains underfunded and underdeveloped, forcing reliance on external multilingual models connected via tools like Whisper.
21 Sep 2026: Idea · Use specialized decision-making models (such as Jett or open-source decision engines) rather than expensive general LLMs to make conditional choices without paying high output token costs.
6 Jul 2026: Open · Current LLM research suffers from context bloat because context is copied forward rather than experiencing natural decay (cessation).
6 Jul 2026: Idea · The output evaluation judge model must be trained on a completely different dataset to prevent bias in self-improvement loops.
6 Jul 2026: Decided · Edge model capabilities require 12GB, 15GB, or 24GB RAM devices to execute larger operational tasks locally.
6 Jul 2026: Idea · Open-source the lower-end Machine World models for free to Sri Lankan businesses.
2 Mar 2026: Idea · Trained and fine-tuned user skills will earn tokens when utilized by others, establishing a token economy capable of acquiring physical goods.
2 Mar 2026: Decided · The system architecture is a native AI backbone where layer-level orchestration manages different AI functions, omni models, and tool calls.
23 Feb 2026: Idea · Utilize AI models supporting Sinhala, such as Gemini, along with OCR to assist in cross-checking and validating Dhamma text data against books.
23 Feb 2026: Idea · Consult [name withheld] to review and verify the complex Abhidhamma concepts modeled in the system.
2 Feb 2026: Decided · OpenRouter does not provide video generation models, requiring external API endpoints and tokens for video workflows.
2 Feb 2026: Decided · The system will utilize Docker containers running MCP (Model Context Protocol) tooling with mapped file systems and injected secret token bundles.
19 Jan 2026: Decided · The open-source LTX video generation model requires high-end GPU hardware (such as RTX 3090) to run locally.
12 Jan 2026: Decided · When an API call returns empty data, the LLM hallucinates responses instead of indicating that no data was found.
15 Dec 2025: Idea · Song meaning prompts must be fine-tuned so that the generated explanations strictly align with Abhidhamma doctrine.
15 Dec 2025: Planned · System access and usage will operate on a token-based cost accounting model.
15 Dec 2025: Idea · Multi-user authentication for the frontend should support Google login via Firebase or OpenRouter.
15 Dec 2025: Idea · Agent skills will be exposed to other agents using MCP (Model Context Protocol) and Agent-to-Agent (A2A) protocols.
8 Dec 2025: Idea · Qwen 4.2 is noted as a strong model alternative.
8 Dec 2025: Idea · Set up a centralized API token pool with a $40 monthly budget and shift to tools like Claude Code, DeepSeek, or Qwen models rather than paying for Cursor seats.
8 Dec 2025: Decided · Cursor locked access to premium models despite a recent payment, claiming token limits were exceeded until the cycle resets on the 22nd.
8 Dec 2025: Decided · Sinhala Unicode text extraction from the PDF fails across multiple OCR solutions and vision models.
8 Dec 2025: Idea · Claude Opus 4.5 is recommended for writing consistent model prompts.
11 Aug 2025: Idea · Developers should prompt LLMs using full, descriptive natural language sentences rather than informal abbreviated text to maintain semantic fidelity.
11 Aug 2025: Idea · Human-in-the-loop intervention should remain available for intermediate steps where AI outputs need manual fine-tuning.
11 Aug 2025: Idea · For cost efficiency or local execution, open-source models running on local Llama can be integrated into the workflow instead of paid token APIs.
11 Aug 2025: Idea · The multi-agent system should achieve over 80% to 90% accuracy once weight adjustments and workflow dynamic fine-tuning are completed.
11 Aug 2025: Idea · Use n8n containerized flow models to automate repetitive development tasks across chained AI agents with specific tooling access.
28 Jul 2025: Idea · Model Context Protocol (MCP) should be used to provide tools to LLMs for tasks such as web browsing and data scraping.
28 Jul 2025: Idea · Speaker 1 proposes funding AI tool subscriptions for the mentees to facilitate multi-model development.
28 Jul 2025: Idea · The website should feature audio-reactive visual models where visual elements dynamically groove and blend based on audio frequency patterns.