🚀 MAST v1.0 — Mast Autonomous System Terminal

Unified AI operator — M4STCLAW v3 + OpenWork v12 + EIGENT v4.1

Working > Perfect · Shipped > Planned · 80% in 3hrs > 100% in 3 days

---

What is MAST?

MAST is Mast's (Anuj's) unified personal AI operator — a Jarvis-style system built on top of OpenCode. It merges:

• M4STCLAW v3 — pentest MCPs, bridge_core (recon/vuln/scheduler/agents), SOUL identity
• OpenWork v12 — 15 hardened MCP servers, semantic cache, 56 API key rotation
• EIGENT v4.1 — lean 6-agent config, NVIDIA NIM integration, working-first philosophy

At a glance:

Layer	Count	Role
MCP servers	21	Tools for shell, files, memory, browser, research, vision, pentest, scheduling, agents
Skills	28	Hot-reloadable workflows for research, GUI, safety, memory, pentest, plugins, media
Providers	11	Free-first cloud routing with SMART_KEY prefix detection
Agents	6	Developer, browser, document, multimodal, pentest, orchestrator
Bridge modules	14	M4STCLAW core logic: memory, recon, vuln, scheduler, safety, voice, agents

What's New in MAST over predecessors

Feature	OpenWork v12	M4STCLAW v3	MAST v1.0
MCP Servers	15	19	21
Providers	7	8	11
NVIDIA NIM	❌	✅ (partial)	✅ full
Mistral	❌	❌	✅ 1B tok/mo free
xAI/Grok	❌	❌	✅
Hardcoded keys	✅ (bad)	✅ (bad)	❌ env-only
SMART_KEY detect	❌	✅	✅
Hinglish chain	❌	❌	✅ Sarvam-M
Pentest chain	❌	✅	✅ NVIDIA primary
Task chains	9	9	11
Skills	22	28	28

---

Quick Start

# 1. Extract MAST.zip somewhere (e.g. C:\MAST)
# 2. Run installer
powershell -ExecutionPolicy Bypass -File install.ps1

# 3. Fill in your API keys
notepad %USERPROFILE%\.config\opencode\config\.env

# 4. Done — run OpenCode in any project
cd C:\your-project
opencode

---

Free Providers — Add These First (₹0 cost)

Provider	Prefix	Signup	Best For
Groq	gsk_	console.groq.com	Speed (315 TPS)
NVIDIA NIM	nvapi-	build.nvidia.com	Code, Pentest, Hinglish
Gemini	AIza	aistudio.google.com	Vision, Reasoning
Mistral	msk-	console.mistral.ai	Writing (1B tok/mo FREE)
Cerebras	csk-	cloud.cerebras.ai	Ultra-fast (1000 TPS)
OpenRouter	sk-or-	openrouter.ai	Kimi K2, MiMo-V2, R1 FREE
SambaNova	UUID	cloud.sambanova.ai	Large models free

SMART_KEY: paste any key as SMART_KEY_1=yourkey — prefix auto-detected.

---

Architecture

MAST/
├── SOUL_MAST.md           ← Agent identity (hot-reload, edit anytime)
├── install.ps1            ← Full Windows installer
├── INSTALL_SKILLS.bat     ← Skills-only quick install
├── mcp_servers/           ← 21 MCP servers
│   ├── llm_fallback.py    ← Brain: 11 providers, 11 task chains, semantic cache
│   ├── memory_mcp.py      ← 3-tier memory (RAM + SQLite + ChromaDB)
│   ├── pentest_mcp.py     ← CEH tools (authorized targets only)
│   ├── task_router_mcp.py ← Auto task classification + routing
│   ├── m4st_agent_mcp.py  ← 6-agent OMO Sisyphus orchestrator
│   ├── scheduler.py       ← Hinglish time parsing + APScheduler
│   ├── coding.py          ← Dev agent bridge
│   └── ...15 more
├── bridge_core/           ← M4STCLAW Python modules
│   ├── smart_brain.py     ← Local brain with LRU cache, TF-IDF recall
│   ├── agents.py          ← Plan→Critique→Refine agent loop
│   ├── recon.py           ← OSINT + network recon
│   ├── vuln.py            ← Vulnerability scanning
│   ├── memory_3tier.py    ← SQLite + ChromaDB memory
│   └── ...9 more
├── skills/                ← 28 OpenCode skills (SKILL.md files)
├── config/
│   ├── .env               ← API keys (NEVER commit this)
│   ├── opencode.json      ← OpenCode config
│   └── m4st_agents_config.json ← 6-agent specs
└── data/                  ← Runtime data (cache, memory, logs)

---

Algorithm

MAST runs a skill-first, safety-gated, task-aware routing loop:

User request
  ↓
SOUL_MAST identity + project context
  ↓
Safety guard
  ├─ blocked if destructive / unauthorized
  └─ allowed if safe
  ↓
Skill search
  ├─ replay known workflow if matched
  └─ route to MCP if no skill hit
  ↓
Task classifier
  ├─ code / research / vision / pentest / hinglish / agent / write / speed
  ↓
Provider chain
  ├─ try best model
  ├─ rotate key on rate limit
  └─ fallback across free-first providers
  ↓
Memory log + response

For a deeper walkthrough, see ALGORITHM.md.

---

Task Chains

Chain	Primary Model	Use Case
speed	Groq → Cerebras	Quick answers, tool calls
reason	Kimi K2 → DeepSeek R1	Math, logic, analysis
code	Kimi K2 → Qwen3-Coder	Write, debug, refactor
vision	Gemini 2.5 Flash → MiMo-Omni	Screenshots, GUI
research	Kimi K2 → DeepSeek R1	Multi-source research
write	Mistral Large → Cerebras	Docs, emails, blogs
agent	Kimi K2 → Qwen3-235B	Multi-step automation
pentest	NVIDIA/deepseek-r1 → NVIDIA/GLM	CEH labs only
hinglish	NVIDIA/Sarvam-M → Gemini	Indic/Hinglish tasks
vision_reason	Gemini 2.5 Flash → MiMo-Omni	Visual reasoning
default	Groq → Cerebras → NVIDIA → Gemini	Fallback

---

Project Resources

Resource	Description
ECOSYSTEM.md	Full M4ST repo map and how the projects fit together
ALGORITHM.md	Routing algorithm, fallback loop, memory flow, and safety gate
PRESENTATION.md	Launch deck-style walkthrough
SOCIAL.md	Tweets, LinkedIn copy, captions, hashtags
DEMO_STORYBOARD.md	GIF/video storyboard for demos
SOUL_MAST.md	Hot-reloadable identity and operating policy
CHANGELOG.md	Fixes and release notes

---

Security

• Safety guard P0: blocks rm -rf, DROP TABLE, format c:, del /f /s
• localhost only: never binds to 0.0.0.0
• Pentest: authorized_targets.txt checked in CODE before any scan
• Keys: .env only — add to .gitignore immediately
• Telegram: MAST_ALLOWED_CHAT_IDS only

---

Adding More Keys

# In config/.env — add as _1, _2, _3...
NVIDIA_API_KEY_1=nvapi-xxxxx
NVIDIA_API_KEY_2=nvapi-yyyyy

# Or use SMART_KEY auto-detect
SMART_KEY_1=nvapi-xxxxx   # auto-detected as NVIDIA
SMART_KEY_2=msk-yyyyy     # auto-detected as Mistral
SMART_KEY_3=gsk_zzzzz     # auto-detected as Groq

---

MAST v1.0 — May 2026 | mastjarvis-cmyk | 80% in 3hrs > 100% in 3 days