Multi-agent code orchestration · BYOK · No token markup

WAKE UP TO
A PR YOU CAN
ACTUALLY MERGE

aiorch runs parallel AI agents on your codebase. They write the code, review each other's work, resolve conflicts, and open a GitHub PR — all while you're away. Your keys, your cost, zero middleman.

RUN YOUR FIRST SESSION → ▶ Watch 2-min demo
5min
Install to first session
N
Parallel agents per session
0
Token markup — your keys, your cost
1PR
Delivered per completed session
▶ What actually happens when you run it
live session log · real output
aiorch · session #2841 · feature/payment-refactor
$ aiorch start "Refactor payment module, add retry logic, write tests"
◆ Task decomposed into 4 sub-tasks
◆ Creating isolated git worktrees...
[agent-1] branch: feat/payment-core — Refactoring PaymentService
[agent-2] branch: feat/payment-retry — Implementing ExponentialBackoff
[agent-3] branch: feat/payment-tests — Writing unit + integration tests
[agent-4] branch: feat/payment-docs — Updating API documentation
⚡ Review round 1/3 · agent-2 needs revision → sent back
⟳ agent-2 · retry attempt · edge case handling improved
✓ Review round 2/3 · all agents approved
◆ Merging branches → integration/payment-refactor
◆ Conflict resolved: src/services/payment.ts (line 142)
✓ Integration tests passed · 38/38
⬆ PR created → github.com/yourorg/yourrepo/pull/847
✓ Session complete · 4 agents · $0.84 (your OpenAI key)
$
Git worktree isolation ·
Parallel agent execution ·
Multi-round code review ·
Automatic conflict resolution ·
GitHub PR delivery ·
Bring your own API keys ·
Claude · OpenAI · Gemini · Ollama ·
Real-time cost tracking ·
Zero token markup ·
Git worktree isolation ·
Parallel agent execution ·
Multi-round code review ·
Automatic conflict resolution ·
GitHub PR delivery ·
Bring your own API keys ·
Claude · OpenAI · Gemini · Ollama ·
Real-time cost tracking ·
Zero token markup ·
The key difference

MOST AI TOOLS
GIVE YOU CODE

WE GIVE YOU A
REVIEWED PR

Every other tool
Generates code. Dumps it on you. You still need to read it, test it, review it, fix it, and open the PR yourself.
aiorch
Agents write code, then a separate reviewer agent critiques it, sends it back, forces revisions, and only merges when it passes. PR lands on GitHub. You just review and click merge.
The code was already reviewed before you saw it. That's the entire product.
What you get

NOT EFFORT.
OUTCOMES.

PR
Ready to merge
A real GitHub PR with summary, diff stats, and agent breakdown. Not a code dump.
✓
Already reviewed
Code went through multiple rounds of AI review before it ever reached you.
⟳
Conflicts resolved
Merge conflicts between parallel agents are automatically detected and resolved.
$
Cost you control
Exact cost per session, per agent. Your keys, your rate. We take zero cut.
Demo

SEE IT LIVE

Watch aiorch take a real task, deploy agents, review the output, and open a GitHub PR — in 2 minutes.

aiorch · demo session
$ aiorch start "Add user authentication..."
◆ Task decomposed into 3 sub-tasks
[agent-1] branch: feat/auth-core — Building JWT middleware
[agent-2] branch: feat/auth-routes — Creating login endpoints
⚡ Review round 1/2 · agent-1 needs revision
✓ Review round 2/2 · all approved
⬆ PR created → github.com/yourorg/repo/pull/312
▶
Watch Demo · 2 min
Task → Agents → Review → GitHub PR
Proof it works

A REAL PR.
CREATED BY AIORCH.

● OPEN Refactor payment module: retry logic, error handling, full test coverage
aiorch-bot opened this pull request · 4 commits · into main from integration/payment-refactor
ai-orchestrated 4 agents review: 2 rounds cost: $0.84 tests: 38/38 ✓
## Summary
This PR was generated by aiorch. Task: "Refactor payment module, add retry logic, write tests"
## Changes
✓  PaymentService refactored with clean separation of concerns
✓  ExponentialBackoff retry logic with configurable limits
✓  38 unit + integration tests (100% coverage on changed files)
✓  API documentation updated
## Agents
agent-1 (claude-opus-4) · agent-2 (claude-sonnet-4) · agent-3 (claude-sonnet-4) · agent-4 (openai:gpt-4o) — review model
Session ID: orch-2841 · aiorch.ai/sessions/2841 · Total cost: $0.84 (your API keys)
6 files changed +284 additions −91 deletions
VIEW ON GITHUB →

This is what aiorch delivers. A PR you can actually review and merge.

How it works

FROM TASK
TO PR

aiorch decomposes your task into parallel workstreams, isolates each agent in its own git branch, runs automated review loops, and merges everything into a clean PR — fully unattended.

01
Decompose
Your task is analyzed and broken into parallel sub-tasks with clear ownership and dependencies.
02
Isolate
Each agent gets its own git worktree and branch. Zero interference between parallel workstreams.
03
Review
Multi-round automated code review. Agents negotiate, revise, and resubmit until approved.
04
Merge
Approved branches merge into an integration branch with automatic conflict resolution.
05
PR
A detailed GitHub PR appears with summary, cost, file breakdown, and review history.
Your keys. Your cost.

ZERO
MIDDLEMAN

aiorch never touches your API spend. Plug in your existing keys — Anthropic, OpenAI, Azure, or run fully local with Ollama. You pay models directly at your negotiated rate.

  • Enterprise contracts with Anthropic or Azure OpenAI? They work out of the box.
  • API keys encrypted locally. Never transmitted to aiorch servers.
  • Per-role model selection — different models for agents, planning, and review.
  • Exact cost tracking per agent per session. No surprises.
ANT
Anthropic Claude
claude-opus-4, claude-sonnet-4
AGENT ✓
OAI
OpenAI
gpt-4o, o3, o4-mini
PLAN / REVIEW
GEM
Google Gemini
gemini-2.0-flash, gemini-2.5-pro
PLAN / REVIEW
AZ
Azure OpenAI
Your enterprise endpoint
PLAN / REVIEW
OLL
Ollama (Local)
llama3, mistral, codestral
LOCAL · $0
🔒 API keys stored encrypted on your machine.
They never leave your infrastructure.
Built for serious codebases

NOT A DEMO.
A PRODUCT.

⎇
Git Worktree Isolation
Every agent works in its own git worktree. No race conditions. No shared state. Agents can't interfere with each other's work, even when editing the same repository.
Core architecture
⟳
Multi-Round Review Loop
Automated reviewer evaluates each agent's output. Agents negotiate, revise, and resubmit. Work doesn't merge until it passes. Configurable rounds and quality gates.
Quality control
⬆
GitHub PR Delivery
Session complete? A PR appears. Includes summary, files changed, diff stats, agent breakdown, cost, and review history. The screenshot your CTO will ask about.
The magic moment
⚡
Live Cost Tracking
Real-time cost per agent, per session. Exact figures for OpenAI, estimates for Claude CLI, $0 for Ollama. Pre-run cost estimates before you start. No bill shock.
Financial transparency
⊕
Pipeline Orchestration
Chain sessions into multi-phase pipelines. Sequential and parallel execution. Complex workflows that span planning, implementation, testing, and documentation — automated end to end.
Advanced workflows
◎
Audit Trail
Every event logged with typed audit records. State machine with atomic writes and file locks. Every decision traceable. Designed for production, not just demos.
Production grade
THE PR IS WAITING WHEN
YOU COME BACK FROM LUNCH
That's the moment developers decide it's worth it
Pricing

SIMPLE.
HONEST.

Developer
$49/mo
For solo developers and indie founders who want to ship faster without hiring.
  • Unlimited sessions
  • Up to 8 parallel agents per session
  • GitHub PR integration
  • All LLM providers (BYOK)
  • Real-time cost tracking
  • 14-day free trial
START FREE TRIAL — 14 DAYS
MOST POPULAR
Team
$99/mo
For technical founders and small engineering teams who need to multiply output.
  • Everything in Developer
  • Unlimited parallel agents
  • Pipeline orchestration (multi-phase)
  • Advanced review hooks + quality gates
  • Secret scanning + security audit
  • Priority support
RUN YOUR FIRST PR FREE →

You pay aiorch for the workflow engine. You pay LLMs directly at your own rate. Zero markup. Always.

START YOUR
FIRST PR

Install in 5 minutes. First PR in 15.
Your keys. Your code. Zero middleman.

RUN YOUR FIRST SESSION →
curl -fsSL aiorch.ai/install.sh | bash ⎘ copy