Talk to the App to Change the App

The seed

Every application is a frozen set of decisions someone else made. When your need falls outside those decisions, the path forward is slow and indirect: describe the gap, file a request, wait for it to be prioritized, wait for it to ship. The person who knows exactly what they want is the person least able to make the software do it.

This quest asks a different question. What if describing the change was the change? An application that listens to plain language about how it should behave, then rewrites its own structure to match — turning the user from a petitioner into an author. Meta-programming, but with conversation as the editing surface instead of code.

Why it matters

Sovereignty over your tools means more than self-hosting them. A tool you cannot reshape still belongs to whoever shaped it. If software can be re-formed by the people who use it, the gap between “what the tool does” and “what I need it to do” stops being a backlog item and becomes a sentence you speak.

• Agency — the user, not a distant team, decides what the software becomes.
• Speed — the loop from need to working change collapses to a conversation.
• Local-first — the reshaping happens on infrastructure the user owns, with no roundtrip to a vendor.

What building it out looks like

• Define a safe, inspectable representation of the app’s behavior that conversation can target — so changes are legible and reversible, not opaque.
• Build the conversational layer that translates a described need into a concrete, scoped modification of that representation.
• Add a review-and-rollback gate so every self-applied change can be seen, confirmed, and undone before it sticks.
• Prototype against a single small application to prove the loop end to end before generalizing it.

Dependencies & lineage

Technically this depends on something like The Integrated LLM System — a context-aware, privacy-respecting personal AI is the substrate that makes conversational self-modification safe — and, to some extent, on the developmental layer behind developmental world news (modeling shifts and transitions). It also touches Trustworthy Servers, where the reshaping runs without a vendor roundtrip.

It connects to an older idea (a Tana note from the Lieboch years, ~2023): that LLMs could accelerate cognitive development by sensing a user’s next developmental step and scaffolding toward it — with heavy data-privacy guardrails, since a system that can map cognitive states is deeply personal (one step removed from the emotional, and mappable by others in any case). The same lineage reached toward open-source neural interfaces (a Neuralink-style implant, but open) as a longer-horizon substrate.

Note: a related effort “coasys” was referenced but could not be located in the local code or in Tana — who is working on this in general is still to be confirmed rather than asserted here.

Status

Dormant. The idea is captured and the shape is clear, but no active build has started. It waits for a concrete first host application to ground the experiment.