LocalPTP — LocalCode Orchestrator

A command-line orchestrator that drives a local coding model (served by LM Studio) over a large codebase. It indexes your repo, builds focused context packages, decomposes a task into ordered subtasks, and applies model-generated diffs through safety gates with approval checkpoints — then reviews and summarizes the work.

Everything runs against a model on your own machine. No code leaves your computer.

Features

• Repo indexing — scans the codebase into .ai-orchestrator/index.json and a /ai memory map.
• Context builder — assembles role-scoped, token-budgeted context for the model.
• Task planning — decomposes a task into ordered subtasks via the model.
• Step/run loop — model → diff → safety gates → approval → apply → tests, one subtask at a time or looped.
• Safety gates — approval prompts, risky-path confirmation, changed-file limits, binary/path-traversal refusals.
• Review — advisory review of the current Git diff (modifies nothing).
• Summarize — closes the loop by updating /ai memory from the session.

Requirements

• Node.js ≥ 20
• LM Studio running locally with a model loaded and its OpenAI-compatible server started (default http://localhost:1234/v1).
  • Default model: qwen/qwen3.6-27b (configurable — see Configuration).
• Git (the orchestrator operates on a Git working tree).

Installation

Install from npm (recommended)

npm install -g localptp
localptp <command>

Try it without installing

npx localptp <command>

Clone and build (for contributors)

# 1. Clone
git clone https://github.com/AlmogMaayan/LocalPTP.git
cd LocalPTP

# 2. Install dependencies
npm install

# 3. Build
npm run build

Run it (contributor workflow)

During development (no build step, runs the TypeScript directly):

npm run localptp -- <command>

After building, run the compiled CLI:

node dist/cli.js <command>

Or install it globally so localptp is on your PATH:

npm install -g .
localptp <command>

Deprecated alias: The localcoder binary resolves to the same CLI for one release and will be removed thereafter. Please update any scripts to use localptp.

Quick start

# 0. Start LM Studio, load qwen/qwen3.6-27b, and start its local server.

# 1. Scaffold /ai memory + .ai-orchestrator/config.yml (idempotent)
localptp init

# 2. Verify LM Studio is reachable and the model responds
localptp doctor

# 3. Index the repo
localptp index

# 4. Create a task
localptp task "add input validation to the signup form"

# 5. Plan it into subtasks (calls the model)
localptp plan

# 6. Execute — one subtask, or loop until a stop condition
localptp step
localptp run

# 7. Review the diff and summarize the session
localptp review
localptp summarize

Commands

| Command | Description | |---------|-------------| | init | Scaffold /ai memory + .ai-orchestrator/config.yml (idempotent; no source edits). | | config [key] [value] | Show the merged config, a sub-tree, or set a dotted key (e.g. config model.model). | | doctor | Verify LM Studio is reachable and the model responds. | | index | Scan the repo and write .ai-orchestrator/index.json + update /ai memory map files. | | context [--role <role>] | Preview the context package for a role (read-only; no model call). | | task <text> | Create a scoped task + session and mark them active. | | plan | Decompose the active task into ordered subtasks (calls the model). | | resume [index] | List past sessions and continue from a selected one. | | step | Run the next pending subtask: model → diff → safety gates → approval → apply → tests. | | run | Loop the step cycle over pending subtasks with approval checkpoints until a stop condition. | | review | Review the current Git diff with the model (advisory; modifies nothing). | | summarize | Summarize the session and update /ai memory files. |

Global flags: --json (structured output) · --debug (verbose logging).

Configuration

localptp init writes .ai-orchestrator/config.yml. Sensible defaults apply when a field is absent. Key model settings:

model:
  provider: lmstudio
  base_url: http://localhost:1234/v1
  model: qwen/qwen3.6-27b
  api_key: lm-studio
  temperature: 0.2
  max_context_tokens: 32768
  timeout_ms: 60000

Read or change values from the CLI:

localptp config model.model              # show current model
localptp config model.model qwen/qwen3.6-27b   # set the model
localptp config model.baseUrl            # show the LM Studio endpoint

Safety, context, and command settings (typecheck/lint/test/build) are also configurable — run localptp config to see the full merged tree.

Development

npm run build       # compile TypeScript to dist/
npm run typecheck   # type-check without emitting
npm test            # run the vitest suite

License

MIT © 2026 Maayan Almog