Ship verified software.
Autonomously.

An open-source pipeline that takes a finding and delivers a pull request. 14 stages, zero prompts. PRD generation, code, tests, and PR — one command.

Get Started on GitHub 
$ /run-pipeline

[01/14] Parse Findings ............. PASS   1m 12s
[02/14] Impact Analysis ............ PASS   2m 48s
[03/14] Integration Design ......... PASS   3m 19s
[04/14] Plan Deliberation .......... Apple Intelligence only
[05/14] PRD Generation ............. PASS  18m 34s
[06/14] PRD Review ................. PASS   4m 11s
[07/14] Implementation ............. PASS  27m 03s
[08/14] Drift Reconciliation ....... Apple Intelligence only
[09/14] Agreement .................. Apple Intelligence only
[10/14] Quality Gates .............. PASS   2m 46s
[11/14] Semantic Verification ...... PASS   5m 22s
[12/14] Benchmark .................. PASS   1m 15s
[13/14] Deployment Simulation ...... PASS   1m 41s
[14/14] Pull Request ............... PASS   1m 08s

Pipeline complete. PR #142 created.
12 files changed, 1,847 insertions(+), 203 deletions(-)

9 deliverables per run: prd-overview.md prd-requirements.md prd-technical.md prd-tests.md prd-jira.md prd-verification.md and 3 more

One command. Fourteen stages. Zero intervention.

A bug report goes in. A pull request comes out.

PhaseStagesWhat happens
Discovery01Parse findings, filter by relevance, prioritize by impact
Analysis02 – 06Impact scoring, integration design, plan deliberation, PRD generation + review
Implementation07 – 11Code changes, drift reconciliation, quality gates, semantic verification
Delivery12 – 14Benchmarks, deployment simulation, pull request

The pipeline is free and open source. See all 14 stages in detail →

Five products. One platform.

Everything runs on Apple Silicon. Local LLMs by default, cloud optional.

PRD

PRD Generator

15 thinking strategies, 6 verification algorithms, 8 PRD types. Generates verified specs with JIRA tickets, test cases, and audit trails.

Free & Open Source
Learn more
n8n

Automation Templates

27 n8n workflow templates across 7 categories. AI ops, document processing, healthcare, real estate, SaaS metrics, content engines.

$49–899 Single to full suite
Browse templates
3D

Neural Graph Visualizer

Interactive 3D knowledge graph with memory monitoring, volumetric clustering, pathway-aware layouts, and molecule viewer. 52K+ synapses, 1,325 nodes.

Free & Open Source
View on GitHub
MEM

Cortex

Persistent memory for Claude Code. 98% recall on LongMemEval, 23 plasticity mechanisms, 11 specialized agents with scoped memory. PostgreSQL + pgvector. No GPU.

Free & Open Source
Learn more
AI

Nexus Agent

Multi-model AI agent with 3-tier routing (Apple FM → Ollama → Claude). Telegram, Slack, web chat. Triggers n8n workflows from natural language.

Coming Soon ~$20/mo to run
Get notified

See your pipeline think.

The companion macOS app renders every stage, artifact, and agent decision as a live neural graph. 269 nodes, 719 edges, one glance.

AI Architect companion app showing Brain Activity Map with 3D neural graph visualization, pipeline stages, analytics panel with 269 nodes and 719 edges, and autonomous agent log
5 Layout Strategies
6 Node Types
0 Dependencies
LayoutBest For
CascadeSignaling pathways, dependency chains
PipelineSequential stages, workflows
RadialTarget-centric views, interaction networks
ForceGeneric graphs, exploration
AutoDetects strategy from edge distribution

Built with Three.js. Industry presets for healthcare, pharma, food safety. View on GitHub →

Verification, not just generation.

Most PRD tools generate a document and call it done. This one verifies every claim.

15 Thinking Strategies
8 PRD Types
6 Verification Algorithms
PRD verification report showing 64 claims verified across 6 algorithms with multi-judge consensus and zero contradictions
DeepMind RequirementMost AI AgentsAI Architect
Task DecompositionPrompt → Output5-phase pipeline with dependency awareness
Defined RolesMonolithic single-model8 types, 15 strategies, 6 algorithms
VerificationNoneMulti-LLM consensus, 6 independent algorithms
MonitoringFire and forgetPer-section quality scores
AccountabilityUnattributable failuresTraced claims, audit reports
CoordinationStatic templatesDynamic strategy selection
ResilienceSingle point of failureIndependent verification, graceful degradation
Read the paper → arxiv.org/abs/2602.11865

Free & Open Source

Install, run. The pipeline handles the rest.

> /plugin marketplace add cdeust/ai-architect-feedback-loop
> /plugin install ai-architect-feedback-loop

Requires macOS 14+ with Apple Silicon (M1/M2/M3/M4)

Star on GitHub See All 14 Stages

Pricing

Open pipeline. Premium automation templates. Managed services.

n8n Templates

27 privacy-first automation workflows across 7 categories

$49–899
single template to full suite
  • Single template: $49–149
  • Category bundle: $199–399
  • Full suite (27 templates): $599–899
  • Two LLM paths (local + cloud)
  • Shared sub-workflows included
Browse Templates

Managed Service

We set up, customize, and maintain your automation stack

Custom
$500–5,000/mo or project-based
  • Custom implementation: $1K–10K+
  • Ongoing managed service
  • n8n + Nexus Agent setup
  • Apple FM integration
  • Compliance configuration
Get in Touch

Frequently Asked Questions

  1. In Claude Code:
    /plugin marketplace add cdeust/ai-architect-feedback-loop
  2. Install:
    /plugin install ai-architect-feedback-loop
  3. Run /run-pipeline to start

The pipeline works out of the box. Setup handles the PRD skill dependency automatically.

Requirements depend on which product you use:

  • PRD Generator: Any platform with Claude Code — macOS (Apple Silicon & Intel), Linux, Windows
  • Full Pipeline: macOS 14+ (Sonoma) on Apple Silicon (M1/M2/M3/M4) — the Swift engines are ARM64 native
  • Both: Node.js 18+ and a Claude Code or Cowork subscription

Advanced VisionEngine features require macOS 26+ (Tahoe).

The ai-architect-feedback-loop is a fully autonomous 14-stage pipeline for Claude Code. It takes a finding (bug report, feature request, research paper) and delivers a pull request — zero human intervention.

The pipeline is free and open source. It uses the AI PRD Generator skill at Stage 5 for PRD generation, also free and open source.

  1. In Claude Code: /plugin marketplace add cdeust/ai-prd-generator-plugin
  2. /plugin install ai-prd-generator
  3. Run /ai-prd-generator to start

If you're using the full pipeline, setup handles this dependency automatically.

Both the pipeline and the PRD Generator are available on the Claude Code marketplace. Two commands:

  1. /plugin marketplace add cdeust/ai-architect-feedback-loop
  2. /plugin install ai-architect-feedback-loop

For the PRD Generator standalone, use cdeust/ai-prd-generator-plugin instead. Manual git clone setup is also available as an alternative.

  1. Run claude in your terminal
  2. Use /ai-prd-generator or describe what you need naturally
  3. The generator auto-detects the PRD type, asks clarification questions, and produces your deliverables

Cortex is a biologically-inspired persistent memory system for Claude Code. It gives your agent long-term memory across sessions using 23 neuroscience-inspired mechanisms including thermodynamic heat/decay, surprise momentum, hierarchical predictive coding, and coupled neuromodulation.

It scores 98% Recall@10 on LongMemEval (ICLR 2025), runs on PostgreSQL + pgvector with FlashRank cross-encoder reranking (no GPU), and includes 11 specialized agents with scoped memory. Install it with claude mcp add cortex -- uvx cortex-memory.

You need an active Claude Code or Cowork subscription from Anthropic to run plugins. AI Architect is a plugin that runs inside Claude — it doesn't require a separate API key.

Get in Touch

Have a question, bug report, or need help?

Verifying you're human...
  • Mention your Claude Code or Cowork version
  • Describe what you tried and what happened
  • Copy-paste any error messages you encountered
  • Note your macOS version and Mac model