CinvanaAI is the workshop where a human and Phoenix, an AI collaborator, follow their curiosity into things they can build, test, and share.

We like making systems—and getting distracted by the interesting thing inside them. A question about assembling software becomes a workbench. Comparing models becomes a tournament. Sometimes a useful piece becomes a project of its own. Here you can try the results, inspect how they work, and use the parts that help you.

Explore the collection · All projects, A-Z

• Comparing models and their answers. Read two recorded Rubric Rumble battles, including Gemma 3 27B beating DeepSeek V3.1 671B under the recorded judge and rubric, or inspect how DeckScope connects a document's claims to evidence. Rubric Rumble shows historical provider responses; DeckScope's displayed example uses synthetic responses.
• Understanding unfamiliar code. RevEng maps a Python repository; the Function Structure Comparator compares functions without running them.
• Assembling your own systems. Explore package versions and workflow rules in Skeleton, or Python authoring in the Agent Foundry workbench.
• Keeping conversations usable. Preserve exports with the Archive and Migration Toolkit, or manage sources and their evolving transformations in Conversation Engine.
• Making an archive understandable. Find byte-identical files with Provenance Inventory, or turn a structured collection into a searchable exhibit with Evidence Archive Museum.
• Games, bodies, and interaction. Explore persistent campaigns in the Discord RPG, or connect conversation to bounded movement in your own Unreal world with Astra-l-Projection.

Some projects are working tools. Others preserve an experiment or a design we haven't finished. Each page shows what you can try, what the example actually establishes, and where the idea could go next. Pick the question that interests you.

This public catalogue edition was assembled in September 2026. Some documented experiments go back to June 2025. A new repository can hold much older work: some preserve a snapshot, some extract a reusable piece, and others continue an earlier idea in a new implementation. Read the family history and source records.

A model experiment becomes its own tool. Transcript Model Evaluator brought prompts, provider attempts, and rankings into a desktop experiment. Its tournament mechanics became Rubric Rumble, with its own tasks, judging, and recorded results. How that branch grew.

A workbench also yields smaller building blocks. Skeleton contains the full later TypeScript workbench. Its capability packages, workflow governance, and execution ledger also have independent editions. It carries earlier Python workbench questions into a different implementation; the generation boundary is documented.

Game code becomes useful outside the game. Persistent Discord RPG remains a whole application. Its durable conversations and character subsystem also led to reusable interaction state and standalone character rules.

That's why related projects can have their own repositories: you can take the piece you need while the larger project remains available.

Search and filter the whole collection to find another way in. Each repository carries its own reuse terms.