Mayo 🦾🤖

The Autonomous Triple-AI Maintainer

Mayo is a Self-Improving Autonomous Maintenance Engine integrated directly into your GitHub ecosystem. It uses a Triple-AI Pipeline — three specialized AI models working in concert — to produce high-value, validated code improvements across all your repositories.

---

🧬 Triple-AI Pipeline

Every improvement goes through 3 AI models before it becomes a PR:

flowchart TD
    A["Hourly Cron Trigger"] --> R0["REVIEWER: Audit pending PR statuses"]
    R0 --> B["SCANNER: Deep codebase analysis"]
    B -->|"Text-only summary + plan"| C["EXECUTOR: Generate surgical edits"]
    C -->|"Proposed search/replace JSON"| D["REVIEWER: Validate edits"]
    D -->|"APPROVE"| E["Create PR"]
    D -->|"CORRECT"| F["Apply corrected edits then Create PR"]
    D -->|"REJECT + feedback"| C2["EXECUTOR: Retry with feedback"]
    C2 --> D2["REVIEWER: Validate retry"]
    D2 -->|"APPROVE"| E
    D2 -->|"REJECT"| SKIP["Skip, save failure to memory"]
    E --> MEM["All 3 AIs save lessons to Global Memory"]
    F --> MEM

Loading

Role	Model	Purpose
🔭 Scanner	Gemini 2.5 Flash	Reads full codebase → text-only analysis (zero compaction risk)
⚡ Executor	Llama 3.3 70B (Groq)	Receives plan → produces surgical search/replace edits
🛡️ Reviewer	Gemini 2.5 Flash	Validates edits, corrects mistakes, audits PR review history

---

🧠 Cross-Repo Global Memory

Unlike standard AI bots, Mayo has persistent memory:

• Tracks successes, failures, and "lessons learned" across all repositories.
• Insights from Repo A directly improve work on Repo B.
• The Reviewer audits real PR states (merged/closed/commented) and updates memory automatically.

---

🩺 Surgical Precision

The Executor uses a Search/Replace block system (max 10 lines per block). This guarantees:

• 100% preservation of your original code structure.
• Zero hallucination of unrelated code.
• Validated PRs — every edit is reviewed by the Reviewer before creation.

---

🏗️ Analysis Depth

The Scanner performs a rigorous multi-layered analysis:

1. Security: Injections, hardcoded secrets, missing validation
2. Logic: Edge cases, null checks, error handling
3. DX: Missing READMEs, build guides, setup docs
4. Performance: Redundant calls, memory leaks
5. Consistency: Naming, patterns, style
6. Creative: Proactive "expert touches"

---

⚙️ Setup & Deployment

Environment Variables

Variable	Purpose
GEMINI_API_KEY	Scanner (Gemini A)
GEMINI2_API_KEY	Reviewer (Gemini B)
GROK_API_KEY	Executor (Llama 3.3 70B via Groq)
APP_ID / PRIVATE_KEY	GitHub App authentication
CRON_SECRET	Hourly trigger authorization

Deployment

1. Deploy as a GitHub App on Vercel.
2. Point webhook to https://your-app.vercel.app/webhook.
3. Install on your repositories.
4. The hourly cron (.github/workflows/cron.yml) handles the rest.

---

ℹ️ Author

Created by Joseph (@HOLYKEYZ). Advanced agentic engineering for autonomous codebase maintenance.

Happy coding! 🚀 (v3.0 — Triple-AI)