MMAT

MMAT, short for Make Me A Thing, is an interactive planning and implementation tool for repository-based software work.

It opens a TUI, asks what you want to build, explores several solution directions with an LLM, asks you to approve the recommended approach, then plans and executes the work in isolated git worktrees before merging validated changes back into your checkout.

Today the execution pipeline is opinionated toward Rust projects because the built-in validation steps run Cargo commands.

What MMAT does

• Starts with a free-form prompt such as a feature request, refactor, or product idea.
• Runs a discovery stage over the current repository and, optionally, external web research.
• Generates multiple candidate solution branches: pragmatic, best-practice, alternative, contrarian, and ambitious.
• Reconciles those branches into one proposal and asks you to approve it or request revisions.
• Builds an implementation plan and runs an architect-style review over that plan.
• Executes implementation items in isolated git worktrees under .mmat-worktrees.
• Validates each item with cargo fmt --all, cargo check, cargo test, and cargo clippy -- -D warnings.
• Runs a final review and, when needed, schedules remediation passes before finishing.

Requirements

• Rust toolchain
• Git repository with a valid HEAD
• An OpenAI-compatible API endpoint
• A sibling checkout of the NAAF repository at ../naaf/main

MMAT uses path dependencies from ../naaf/main, so this repository is not currently self-contained.

Configuration

MMAT reads these environment variables:

Variable	Purpose	Default
OPENAI_API_KEY	API key for the LLM endpoint	lm-studio
OPENAI_BASE_URL	Base URL for an OpenAI-compatible API	http://127.0.0.1:1234/v1
OPENAI_MODEL	Model name used for all workflow stages	essentialai/rnj-1
OPENAI_ORG_ID	Optional OpenAI organisation id	unset
MMAT_WEB_SEARCH_URL	Optional web search endpoint	unset
MMAT_WEB_SEARCH_API_KEY	Optional API key for the web search endpoint	unset
WEB_SEARCH_URL	Fallback for MMAT_WEB_SEARCH_URL	unset
WEB_SEARCH_API_KEY	Fallback for MMAT_WEB_SEARCH_API_KEY	unset

If no web search URL is configured, MMAT still runs, but external research is disabled.

Build

Clone this repository and ensure the NAAF checkout exists at ../naaf/main, then build normally:

cargo build

For a release build:

cargo build --release

Usage

MMAT uses the current working directory as the project root it will inspect and modify.

If you run cargo run inside this repository, MMAT will operate on this repository. If you want to use MMAT on another project, run the built binary from inside that other repository.

During development, the simplest way to launch it is:

cargo run

To capture TUI debug logging, pass a log path after the Cargo separator:

cargo run -- --debug-log target/tui-debug.log

The TUI will open with the prompt:

What are we building?

Enter a request such as:

Add a CLI flag to export the generated plan as JSON.

From there MMAT will:

1. Inspect the repository and build a discovery brief.
2. Explore multiple solution branches.
3. Present a reconciled proposal and ask for approval or revisions.
4. Plan the implementation.
5. Execute validated changes.
6. Leave the merged result in your working tree.

Operational Notes

• MMAT creates temporary worktrees in .mmat-worktrees while it is implementing tasks.
• It copies the current workspace state into those worktrees, so uncommitted local changes are part of the working context.
• The implementation pipeline assumes Cargo is available and that cargo fmt, cargo check, cargo test, and cargo clippy are meaningful for the target repository.
• The interface is currently TUI-only. The only CLI flag today is --debug-log <PATH> for low-level TUI event logging; there are still no flags for passing the prompt or selecting a target directory.

Development Checks

Before committing changes in this repository, run:

cargo fmt --all
cargo clippy -- -D warnings
cargo test

Horizon

• Track ideas and bugs over time
• Allow injecting ideas ad hoc with a constant development loop
• Ability to feed in resources over time - research articles, code snippets etc.
• Ability to automatically parallelise development using worktrees with a reconcile step to rebase back onto active branch