rdn-swarm · SDLC orchestration · v1

An AI-driven
software factory.

A multi-agent SDLC orchestrator that coordinates Analysts, Project Managers, Developers, and QA agents across multiple LLM providers to iteratively build and refine complete applications. Six feature-parity platforms. Four CLIs, a browser-based control plane, and a native Android client that runs the entire pipeline on-device.

6
Platforms
9
Roles (Mobile)
4
Providers
54
Models
5
Quality Dims
swarm --mode sdlc

PHASE 1 · ANALYST

✓ 4 components · 12 acceptance criteria

PHASE 2 · PM

✓ 4 non-overlapping work assignments

PHASE 3 · DEVELOPERS ×4

· index.html

· styles.css

· game.js

· tests/*.js

PHASE 4 · INTEGRATION

✓ 2 issues auto-fixed

PHASE 5 · QA ×3 · interactive

⚠ 4 critical · 2 major · 7 minor

PHASE 6 · FEEDBACK

Score: 82% / threshold 80%

PHASE 7 · SUMMARY

✓ docs/summary.md

Total cost $0.124anthropic · openai · gemini · deepseek
ProvidersAnthropic9 models·OpenAI35 models·Google8 models·DeepSeek2 models·54 total

Six Platforms

One pipeline. Six surfaces.

Four CLI ports, a browser-based control plane, and a native Android client. Each is feature-complete on the same 7-phase pipeline (9 roles on Mobile), with equivalent concurrency models.

Node.js

Canonical · ~4,040 LOC

  • Promise.all / Promise.race
  • @anthropic-ai/sdk + provider SDKs
  • Reference implementation

Python

asyncio · ~3,000 LOC

  • asyncio.Semaphore + gather
  • Async LLM clients across 4 providers
  • Playwright for visual QA

Go

Goroutines · ~3,590 LOC

  • Channels as semaphores
  • chromedp for browser automation
  • Single static binary

.NET 10

Task.WhenAll · ~3,030 LOC

  • SemaphoreSlim limits
  • HttpClient + System.Text.Json
  • Playwright.NET

Next.js Web

SSE Control Plane

  • Live SSE dashboard
  • Event-sourced replay
  • OIDC via Rdn.Identity

Mobile · Android

Kotlin · 9-role · On-Device

  • Kotlin Coroutines + StateFlow
  • AES-256-GCM key vault
  • Five-dim Quality Scorecard
Tour the mobile app →

The Pipeline

Seven phases. Iterative until done.

Each phase uses specialized agents with role-specific prompts and configurable LLM providers. If the QA score is below the threshold, the pipeline loops back to Developers with the deduplicated feedback.

01

Analyst

Decomposes requirements

02

PM

Assigns non-overlapping work

03

Developers

Parallel code generation

04

Integration

Cross-file consistency check

05

QA

Independent review · all files

06

Feedback

Score + deduplicate

07

Summary

User-facing documentation

score < threshold → loop back to PM with QA feedback

Multi-Provider

Mix every major model, by role.

Different agents have different needs. The Analyst wants cheap and fast. The PM wants reasoning. Developers want code quality. QA wants vision. Swarm lets you assign the right model from the right provider to each role, with real-time streaming, full cost tracking, and per-provider token attribution.

Model Catalog54
9
Anthropic
35
OpenAI
8
Google
2
DeepSeek
cost-by-provider.txt● live
Provider
Calls
Tokens
Cost
anthropic
12
44,405
$0.0678
openai
6
28,100
$0.0342
deepseek
4
18,200
$0.0126
google
4
15,670
$0.0099
TOTAL
26
106,375
$0.1245

New · Mobile

The whole pipeline,
in your pocket.

Swarm Mobile is a native Kotlin / Jetpack Compose Android client that runs the full 9-role SDLC pipeline directly on-device against your own LLM provider keys. AES-256-GCM hardware-backed keystore, sandboxed in-app preview for visual QA, a five-dimension Quality Scorecard, and a bundled benchmark template suite. No backend, no account.

Tour the Mobile App
9
Roles
4
Providers
54
Models
5
Quality Dims
0
Servers
Swarm Mobile home tab

Dive Deeper

Read the architecture. Browse the features.