Transmute

Contents

• Overview
• How Transmute Differs from Compose
• Running a Transmute Job
  • Web app
  • CLI
• The Target Description
• User Ethos
• Output

---

Overview

Transmute is a direct re-implementation workflow. It takes all the phase output files from a previously-decomposed project and feeds them as a single prebuilt prompt into the Compose engine — bypassing the inventory basket entirely.

Use Transmute when you want to re-implement a whole project in a different language or framework and don’t need to cherry-pick individual ideas from the inventory first.

---

How Transmute Differs from Compose

	Compose	Transmute
Input	Individual inventory items from any project(s)	All phase outputs (01–06) from one decomposed project
Selection	Manual basket — pick exactly what you want	Automatic — the whole spec
Cross-project mixing	Yes	No
Best for	Synthesising new designs from reusable ideas	Re-implementing an existing project in a new stack

Both workflows run through the same Compose pipeline and produce the same kind of output file. The difference is how the prompt is assembled.

---

Running a Transmute Job

Web app

Transmute is accessible in two ways:

From the Transmute screen:

1. Navigate to Transmute in the sidebar.
2. Select a source project from the dropdown (any previously-decomposed project).
3. Enter a target re-implementation description — e.g. Re-implement in Rust using Axum and Tokio, keeping the same domain model.
4. Choose a backend and click Run Transmute.

From the Projects screen:

1. Navigate to Projects.
2. Click the Transmute button on any project card. This pre-fills the project name in the Transmute form.

CLI

Transmute is not yet exposed as a dedicated CLI command. To achieve a similar result from the CLI, use the Compose command with --project to include all items from that project’s inventory:

otx compose --output rust-rewrite \
  --project my-original-project \
  --description "Re-implement in Rust using Axum and Tokio" \
  --technology "Rust, Axum, Tokio, SQLx"

This CLI approach uses the inventory items (Phase 6 output) rather than the full phase spec. For a full-spec transmute with phases 01–06, use the web app.

---

The Target Description

The target description is the primary steering input. Be specific about:

• The language and key frameworks — e.g. Rust with Axum not just Rust
• Any architectural constraints that differ from the original — e.g. stateless, no global registry
• What to keep vs. discard from the original — e.g. preserve the retry policy and circuit breaker; replace the custom DI container with standard constructor injection

Good example:

Re-implement in Go using the standard library's net/http and database/sql.
Preserve the event sourcing model and all invariants from the original.
Replace the actor-based concurrency with goroutines and channels.
The output should be deployable as a single statically-linked binary.

---

User Ethos

The user ethos setting (configured in Settings in the web app, or via otx settings --user-ethos in the CLI) is applied to Transmute runs in the same way as Compose runs — injected as a top-level authoritative instruction.

---

Output

Results are saved to Output/Composition/<name>/compose-output.md.

From there, you can feed the output into the Implement pipeline to generate working code.