Parton

AI development
orchestration.

Describe a feature in plain English. Parton plans, parallelizes, and ships it — with tests, commits, and a PR.

Early alpha v0.1.0 — not recommended for production use yet. Share feedback
terminal
$
Feature request

"add invoice export"

Planner

Decomposes the feature into stories, assigns coordination domains, orders into waves

identified 6 stories across 4 domains (Schema, Domain, Api, Ui, Tests)
schema must run first — api and ui depend on invoices table
csv, pdf, api are independent — safe to parallelize
tests run last to verify the full export flow
Wave 0
lane-schemaSchema

Create invoices table, add foreign keys to users

queued
Wave 1parallel
lane-csvDomain

CSV serialization with configurable delimiters

queued
lane-pdfDomain

PDF generation with company branding template

queued
lane-apiApi

POST /invoices/:id/export endpoint

queued
Wave 2
lane-uiUi

Export dialog with format picker and download

queued
Wave 3
lane-testsTests

Integration tests for export flow end-to-end

queued
PR ready

6 commits · typecheck, lint, tests passed · branch ready for review

Install

One command. That's it.

Works on macOS and Linux.

terminal
$curl -fsSL https://parton.run/install.sh | sh

AI coding today is powerful but chaotic.

Prompt chaos

Every request starts from scratch. You re-explain context, re-describe architecture, and hope the AI remembers what you said three prompts ago.

Context overload

Large codebases exceed context windows. The AI sees fragments, misses dependencies, and produces changes that break things it never knew existed.

No coordination

Multiple AI sessions edit the same files with no awareness of each other. Merge conflicts, duplicated work, and silent regressions pile up.

AI with the right context.

Most AI tools send your entire repository to the model — or rely on approximate search. Parton analyzes the repo graph first and builds a targeted context for every task.

Traditional AI coding
Feature request
entire repository
AI model
Too much irrelevant code
Wasted tokens
Hallucination risk
Misses real dependencies
Parton
Feature request
Repo graph analysis
Touch set prediction
src/invoices/export.ts
src/lib/pdf.ts
src/api/invoices.ts
prisma/schema.prisma
src/types/invoice.ts
targeted context
AI worker
Only relevant files & symbols
Repository-aware dependencies
Reduced hallucinations
Better model reasoning

Waves & Lanes.

Waves run one after another. Lanes within a wave run side by side.

start
time →
Wave 0
lane-schema
sequential
Wave 13 lanes · parallel
lane-csv
lane-pdf
lane-api
sequential
Wave 2
lane-ui
sequential
Wave 3
lane-tests
parallel lanes
sequential waves

Parton understands your repository.

Before executing any changes, Parton indexes your codebase structure, dependencies, and patterns — so every AI worker operates with full context.

Dependency graph analysis

Parton builds a full dependency graph of your codebase — modules, imports, exports — so the planner knows which files depend on which.

Touch set prediction

Before execution, each lane declares its predicted file touch set. The scheduler uses this to detect conflicts and prevent parallel edits to the same files.

Dependency cone analysis

For any given change, Parton computes the dependency cone — all files that could be affected — ensuring AI workers receive the right context.

Safe parallel execution.

The scheduler detects conflicts, coordinates AI work, and automatically serializes risky operations — so parallel execution is always safe.

Scheduling rules
Schema migration
SerializedOther lanes depend on the new table structure
CSV export + PDF export
ParallelIndependent modules, no shared files
API route + UI component
ParallelDifferent domains, no file overlap
Two API routes
SerializedSame domain — may share routing or middleware files
Integration tests
Last waveMust verify all prior changes together

Every lane runs in an isolated git worktree. Changes are only merged after validation passes — typecheck, lint, and tests.

Four commands. That's it.

From install to shipping features — Parton stays out of your way.

1

Install

terminal
$ curl -fsSL https://parton.run/install.sh | sh
▸ Detecting platform: darwin-arm64
▸ Downloading parton v0.8.2...
▸ Installing to /usr/local/bin/parton
✔ parton v0.8.2 installed successfully
2

Analyze

terminal
$ parton analyze
▸ Scanning repository structure...
▸ Detected: TypeScript · Next.js · Prisma · TailwindCSS
▸ Indexed 347 files, 28 modules, 12 API routes
✔ Repo intelligence cached in .parton/
3

Run

terminal
$ parton run 'add invoice export'
▸ Planning feature: add invoice export
▸ Created 6 user stories across 3 waves
▸ Wave 1: 3 lanes running in parallel
▸ Wave 2: 2 lanes (waiting for wave 1)
▸ Wave 3: 1 lane (integration tests)
✔ All 6 stories passed validation
✔ PR #142 opened → feature/invoice-export
4

Watch

terminal
$ parton dashboard
▸ Dashboard running at http://localhost:4400
▸ Active runs: 2 | Queued: 1 | Completed today: 7
▸ Cost today: $4.32 | Avg duration: 8m 24s

See it in action.

A real execution run — waves execute in order, lanes run in parallel, and every story is validated before committing.

Run #104— add invoice export
Running
Stories: 6Waves: 3Duration: 4m 12sCost: $1.84
Wave 1
Lane ACompleted
US-001 · DB schema
Lane BCompleted
US-002 · API routes
Lane CCompleted
US-003 · Auth middleware
Wave 2
Lane ARunning
US-004 · List endpoint
Lane BRunning
US-005 · Detail endpoint
Wave 3
Lane AWaiting
US-006 · Integration tests

Run development at any scale.

Parton is free for individual use and always will be. Paid plans add coordination, visibility, and infrastructure for teams.

Free

$0

Everything you need to run Parton locally.

  • Local CLI
  • Repo analysis
  • Dependency graph
  • Touch set prediction
  • Targeted context
  • Wave/lane execution
  • Parallel AI execution
  • Local artifacts
  • Branch output
Soon

Team

$20/dev/mo

Coordinate AI-driven development across your team.

  • Everything in Free
  • Shared run registry
  • Conflict detection
  • Team scheduler
  • Project dashboard
  • Run history & analytics
  • Team visibility
  • Role-based access
  • Hosted infrastructure
Soon

Enterprise

Custom

For organizations that need control, compliance, and scale.

  • Everything in Team
  • Distributed orchestration
  • Advanced scheduler policies
  • Enterprise policy controls
  • Private infrastructure
  • Security & compliance
  • Dedicated support
  • SLA
  • Advanced analytics
  • Org governance

Get early access to Pro features.

Team and Enterprise plans are coming soon. Join the waitlist and we'll notify you when they launch.

FeatureFreeTeamEnterprise
Local CLI
Targeted context
Parallel execution
Team coordination
Shared run registry
Conflict detection
Remote workers
Enterprise policy engine
Private infrastructure

What makes Parton different.

Not another AI code generator — a structured execution engine built for real-world codebases.

AI planning

Automatically decomposes features into right-sized user stories with dependency ordering — no manual breakdown needed.

Parallel execution

Independent stories run simultaneously across isolated git worktrees, cutting feature delivery time by 3–5×.

Repo intelligence

Parton indexes your codebase — file structure, dependencies, patterns — so every AI worker has the right context.

Conflict prevention

A scheduler tracks file touch-sets across lanes and prevents concurrent edits to the same files.

Transparent execution

Every decision, plan, and validation result is logged and visible in the dashboard — no black-box AI.

Get in touch.

Questions about Parton? Enterprise needs? We'd love to hear from you.

Start running development with AI.

Describe a feature. Parton plans, parallelizes, and ships it.