Atabey

БесплатноНе проверен

Atabey — AI Governance & Multi-Agent Platform / Orchestrator — Core Engine

автор: ysfbkr

GitHub Embed

Описание

Atabey — AI Governance & Multi-Agent Platform / Orchestrator — Core Engine

README

Language / Dil: English · Türkçe

Pre-alpha · v0.0.25 · Single-developer project · 548 tests passing (80 files)

License: AGPL-3.0 npm MCP

Agent Atabey is an AI Governance & Multi-Agent Platform / Orchestrator built on MCP (Model Context Protocol). It sits on top of your existing AI coding interfaces — Claude Code, Gemini CLI, Cursor, Codex, Grok, Antigravity, or local LLMs — and turns them into a governed, multi-agent software engineering system.

Layer	Role
Your AI IDE/CLI	Writes code, runs tools, executes shell commands
Atabey MCP Server	39 tools · multi-layer governance (10+ checks) · orchestration · memory
@manager + 13 specialists	Delegation, quality gates, risk control, audit trail
`.atabey/` brain hub	Constitution, memory, knowledge, registry, observability

[!NOTE] Project Status (Pre-1.0 / Active Open-Source Experiment): Atabey is currently in pre-alpha (v0.0.x) and is actively being developed as an open-source experiment by a single developer. While the core features (TF-IDF routing, regex PII masking, deterministic rules) are functional, it is not yet production-ready for enterprise environments.

⚠️ Before using or deploying, please review BLINDSPOTS.md for a detailed list of known limitations, security sandboxing issues, and the development roadmap.

Philosophy: "Order from Chaos"

What Atabey is: A deterministic, rule-based governance and multi-agent orchestration middleware for AI coding assistants.

What Atabey is not: A standalone LLM runtime (unlike LangGraph/CrewAI) or an AI/ML-driven threat analysis platform. Your connected AI assistant generates the code; Atabey disciplines, routes, validates, and coordinates it using deterministic rule-matching.

🚀 How It Works

Atabey connects to your AI interface as an MCP tool server. Once connected, you use @agent commands directly in your AI chat:

You (@backend): "Create a user login service with JWT authentication"
     │
     ▼
  Atabey MCP Server (atabey-mcp/)
     │  Intercepts @agent command
     │  Routes via RoutingEngine (TF-IDF + Semantic)
     │  Routes to @backend specialist
     │  Multi-Layer Governance Pipeline validates
     │  Quality Gate checks output
     │  Stores in Vector Memory
     │
     ▼
  Returns governed, reviewed, audited code

No separate terminal needed. No CLI commands for daily use. Just chat with your AI and use @agent syntax.

🧠 How Atabey Relates to Your AI

[!IMPORTANT] Atabey does not write code. The code is written by the AI assistant you are connected to (such as Claude Code, Gemini CLI, or Cursor). Atabey wraps your AI assistant with a powerful, deterministic software engineering discipline and governance layer:

Routing: Directs instructions to the correct specialized virtual agent profile (@backend, @security, etc.).

Risk Engine: Automatically scores operations for safety and halts risky actions (like mass deletions) for human approval.

Quality Gate: Reviews the generated output against corporate standards, syntax correctness, and test metrics.

Memory: Keeps track of past architectural decisions and specialty conventions so your AI assistant learns from history.

We do not mimic the LLM; we discipline it.

📋 Table of Contents

Platform Architecture
Orchestration Flow
Quick Start
MCP Connection (IDE/CLI)
5 Core Capabilities Overview
How Atabey Plugs Into Your AI
Supported Platforms
Installation
Profile-Based Setup
13 Specialized Agents
39 MCP Tools
7 Core Skills
3-Layer Memory System
Knowledge Base (30+ Standards)
Multi-Layer Governance Pipeline
Core Features
Dashboard
CLI Reference
Architecture
Security
KVKK/GDPR Compliance
EU AI Act Alignment
Token Economy & FinOps
Contributing

Platform Architecture

┌─────────────────────────────────────────────────────────────────────────┐
│                         USER / DEVELOPER                                 │
│              @manager planla · @backend API yaz · @quality audit       │
└───────────────────────────────┬─────────────────────────────────────────┘
                                │
        ┌───────────────────────┼───────────────────────┐
        ▼                       ▼                       ▼
┌───────────────┐      ┌───────────────┐      ┌───────────────┐
│ Claude Code   │      │ Gemini CLI    │      │ Cursor IDE    │
│ .claude/      │      │ .gemini/      │      │ .cursor/      │
└───────┬───────┘      └───────┬───────┘      └───────┬───────┘
        │                      │                      │
        └──────────────────────┼──────────────────────┘
                               │ MCP (stdio) — IDE default
                               ▼
┌─────────────────────────────────────────────────────────────────────────┐
│                    ATABEY MCP SERVER (atabey-mcp)                        │
│  Governance Pipeline · Hermes Queue · Orchestrator · Vector Memory       │
│  Risk Engine · FinOps · Auto-Rollback · Human-in-the-Loop               │
└───────────────────────────────┬─────────────────────────────────────────┘
                                │
                                ▼
┌─────────────────────────────────────────────────────────────────────────┐
│  .atabey/  Brain & Memory (SSOT)     │  .agents/  Unified Agent Hub    │
│  ATABEY.md · memory/ · knowledge/     │  7 platform mirrors + skills    │
└─────────────────────────────────────────────────────────────────────────┘

Monorepo layout:

Package	Purpose
`packages/atabey`	CLI, adapters, agent export, init
`packages/atabey-mcp`	MCP server, governance middleware, dashboard API
`packages/shared`	Shared constants, PII, audit, storage (SSOT)

Orchestration Flow

sequenceDiagram
    participant U as Developer
    participant AI as AI Assistant
    participant MCP as Atabey MCP
    participant M as @manager
    participant S as Specialists
    participant G as Governance

    U->>AI: @manager Plan and deploy Auth module
    AI->>MCP: get_framework_status / read_project_memory
    MCP->>G: PII mask · risk scan
    MCP-->>AI: Phase, trace, agent states

    AI->>MCP: send_agent_message → @backend
    U->>AI: @backend Write JWT service
    AI->>MCP: write_file (governed)
    MCP->>G: Quality gate · auto-rollback snapshot
    MCP-->>AI: Approved output

    AI->>MCP: orchestrate_loop
    MCP-->>AI: Hermes queue processed

    AI->>MCP: send_agent_message → @quality AUDIT
    MCP->>G: Compliance · lint
    MCP-->>AI: APPROVED / REJECTED

Orchestrator auto-start (default): When your IDE connects to the MCP server (mcp.json), Atabey automatically boots the headless AgentLoop orchestrator in the background. No separate terminal is required for daily multi-agent work.

# Optional — interactive TUI dashboard loop
npx atabey orchestrate

# Optional — unified HTTP server + dashboard UI
MCP_TRANSPORT=unified MCP_PORT=5858 npx atabey-mcp

Disable auto-start: set "orchestrator": { "autoStart": false } in .atabey/config.json or ATABEY_AUTO_START_ORCHESTRATOR=false in mcp.json.

Project kinds

Kind	Where	App paths (`apps/backend`, `apps/web`)
consumer	Your application repo after `atabey init`	✅ Scaffolded by init
framework-monorepo	This GitHub repository (`packages/*`)	❌ Not applicable — use `packages/atabey`, `packages/atabey-mcp/dashboard`

Quick Start

1. Initialize Atabey in Your Project

Consumer projects: atabey init scaffolds apps/backend, apps/web, docs/, etc. The Atabey framework monorepo itself does not contain apps/* — those paths appear only after you install Atabey in your own application repository.

# Single platform (your app repo)
npx atabey init gemini --profile freelancer --yes

# All 7 platforms (recommended for teams)
npx atabey init gemini --unified --profile team --yes

# Framework monorepo maintainers only (this repository)
npm install          # postinstall runs atabey:setup if .atabey/ is missing
npm run atabey:setup # or run manually

2. Verify MCP & Platform Health

npx atabey check     # MCP transport, unified layout, AL registry, compliance
npx atabey status    # Phase, trace, agent states

3. Connect to Your AI Interface (MCP auto-starts)

atabey init generates mcp.json with MCP_TRANSPORT=stdio and ATABEY_AUTO_START_ORCHESTRATOR=true. Your IDE spawns the MCP server automatically — orchestrator included. Point your AI assistant to it:

Claude Code:

{
  "mcpServers": {
    "atabey": {
      "command": "npx",
      "args": ["-y", "atabey-mcp"],
      "env": {
        "MCP_TRANSPORT": "stdio",
        "ATABEY_PROJECT_ROOT": "/path/to/your/project",
        "ATABEY_AUTO_START_ORCHESTRATOR": "true"
      }
    }
  }
}

Gemini CLI:

gemini config set mcpServers.atabey.command "npx"
gemini config set mcpServers.atabey.args "[\"-y\", \"atabey-mcp\"]"
gemini config set mcpServers.atabey.env "{\"MCP_TRANSPORT\": \"stdio\", \"ATABEY_PROJECT_ROOT\": \"/path/to/your/project\"}"

Cursor (.cursor/mcp.json — auto-generated by init):

{
  "mcpServers": {
    "atabey": {
      "command": "node",
      "args": ["node_modules/atabey-mcp/dist/atabey-mcp/src/mcp/index.js"],
      "env": {
        "MCP_TRANSPORT": "stdio",
        "ATABEY_PROJECT_ROOT": "/path/to/your/project",
        "ATABEY_AUTO_START_ORCHESTRATOR": "true"
      }
    }
  }
}

[!IMPORTANT] MCP transport modes

stdio (default) — Claude, Cursor, Gemini, Codex, Grok. Set by atabey init and atabey mcp install.

unified — HTTP/SSE server on port 5858 for dashboard and remote clients: MCP_TRANSPORT=unified MCP_PORT=5858 atabey-mcp

Repair a broken config: atabey mcp install

MCP Connection (IDE/CLI)

Step	Command	Result
Init	`atabey init gemini --unified --yes`	Agents + 7 adapter MCP configs + shims
Repair	`atabey mcp install`	Refreshes root `mcp.json` (stdio + auto-orchestrator)
IDE connect	Open Claude / Cursor / Gemini	MCP + orchestrator start automatically
Debug	`atabey mcp start`	Manual stdio MCP for troubleshooting
Health	`atabey check`	Validates `MCP_TRANSPORT`, unified layout, AL registry

Platform orchestration strength:

Platform	Orchestration	Best role
Claude Code	Full (7 skills)	Primary orchestrator
Gemini CLI	Strong (6 skills)	Commander / strategist
Local LLM	Strong	Offline / private
Antigravity CLI	Strong	Custom agent JSON
Cursor / Grok / Codex	Limited (4 skills)	Implementer / editor

4. Start Using in AI Chat

Open your AI interface and simply type:

@backend Create a REST API for user management with CRUD operations
@security Audit the authentication middleware
@quality Run compliance check on the new feature

That's it. Atabey handles routing, quality gates, memory, governance, and audit automatically.

5 Core Capabilities Overview

Atabey AL is built on 5 core capabilities that work together seamlessly:

#	Capability	Description	Score
🛠️	39 MCP Tools	File system, search, messaging, governance, memory, quality, network, orchestration	82/100
🧠	3-Layer Memory	Vector Memory (TF-IDF/OpenAI), Project Memory, Specialty Memory (agent learning)	78/100
🤖	13 Specialized Agents	3-tier hierarchy (Supreme/Core/Recon) with state machine	80/100
🎯	7 Core Skills	Platform-adaptive skills for 7 different AI platforms	72/100
📚	30+ Knowledge Standards	Governance, security, architecture, compliance, deployment standards	85/100
🎯	Overall	AI Governance & Multi-Agent Platform / Orchestrator	74/100

Scores reflect pre-1.0 maturity: strong governance and MCP tooling; orchestration depth varies by platform adapter.

How Atabey Plugs Into Your AI

┌──────────────────────────────────────────────────────────────────┐
│                   YOUR AI INTERFACE                                │
│   Claude Code · Gemini CLI · Cursor · Codex CLI · Local LLM      │
│   You type: @backend Create login service                         │
├──────────────────────────────────────────────────────────────────┤
│                    MCP PROTOCOL (stdio/SSE)                        │
│   JSON-RPC messages over stdin/stdout or HTTP/SSE                 │
├──────────────────────────────────────────────────────────────────┤
│                    ATABEY MCP SERVER                               │
│   ┌──────────────────────────────────────────────────────────┐   │
│   │  39 Tools · Multi-Layer Governance Pipeline ·               │   │
│  │  Risk Engine · Loop Detection · FinOps · Auto-Rollback  │   │
│   └──────────────────────────────────────────────────────────┘   │
├──────────────────────────────────────────────────────────────────┤
│                    GOVERNANCE LAYER                                │
│   Quality Gate → Risk Engine → Vector Memory → Audit Log         │
│   PII Masking → Discipline → License Scan → Injection Protection │
└──────────────────────────────────────────────────────────────────┘

Atabey is not a separate execution engine. It is a context-aware governance and policy middleware that intercepts, validates, and routes the actions of AI coding assistants:

13 Specialized Agent Contexts (injected templates to structure AI reasoning)
Deterministic Quality Gates (AST analysis + lint + governance validation)
Risk Gate & Heuristic Scanning (blocking dangerous commands, requiring human approval)
Persistent Vector Memory (TF-IDF + OpenAI embeddings, cosine similarity search)
Audit trails (every action logged, supporting KVKK/GDPR technical alignment)

Supported Platforms

Atabey AL supports 7 platforms with automatic adapter configuration:

Platform	MCP Mode	Agents Export	Tools	Skills	Shim
Claude Code ⭐	`.mcp.json` + `claude_desktop_config.json`	`.claude/agents/*.md`	20 tools	7 skills	`CLAUDE.md`
Gemini CLI	`.gemini/mcp.json`	`.gemini/agents/*.md`	14 tools	6 skills	`GEMINI.md`
Cursor IDE	`.cursor/mcp.json`	`.cursor/rules/*.mdc`	9 tools	4 skills	`CURSOR.mdc`
Codex CLI (Copilot)	`.vscode/mcp.json` + `.mcp.json`	`.agents/instructions/*.md`	9 tools	4 skills	`copilot-instructions.md`
Antigravity CLI	`.agents/mcp_config.json`	`.agents/agents/*.md` + `agent.json`	14 tools	6 skills	`AGENTS.md`
Grok	`.grok/mcp_config.json`	`.grok/agents/*.md`	9 tools	4 skills	`GROK.md`
Local LLM (Ollama)	`.atabey/mcp_config.json`	`.atabey/agents/*.md`	14 tools	6 skills	`LOCAL_AI.md`

Unified Mode: atabey init claude --unified exports agents to ALL platforms simultaneously.

Installation

Requirements

Requirement	Version
Node.js	>= 18.0.0
npm	>= 9.0.0
AI Interface	Claude Code, Gemini CLI, Cursor, Codex CLI, Grok, Antigravity, Local LLM

Quick Install

# Install globally (recommended)
npm install -g atabey

# Initialize with your preferred platform and profile
npx atabey init gemini --profile freelancer --yes

# Verify installation
npx atabey status

# (Optional) Open the web dashboard
npx atabey dashboard

Native Dependency & CI/CD Considerations

Atabey utilizes better-sqlite3 for local state persistence and high-performance vector operations. Since better-sqlite3 includes native C++ modules, it requires binary compilation during installation under certain environments.

If you encounter installation or test run issues (especially in restricted sandbox or CI/CD environments with no internet access or build tools):

Prebuilt Binaries: npm will automatically try to download prebuilt binaries for your platform. Ensure your environment has outgoing network access to the GitHub/npmjs package registries.
Build Tools: Ensure you have python, make, and a C++ compiler (like gcc/g++) installed on your system if you need to build from source.
CI/CD Environments: For CI/CD test runners, you can bypass native sqlite3 compile-time issues by running with the --ignore-scripts flag if you only need the static analysis and typescript compilation checks:
```
npm install --ignore-scripts
```

Profile-Based Setup

Choose the profile that matches your team size. You can also customize your setup using --focus and --lang options.

`--profile freelancer` (1-3 people)

npx atabey init gemini --profile freelancer

Feature	What You Get
Agents	`@manager`, `@quality` + focus-specific agents
Setup Time	~10 seconds
Daily Workflow	Chat with AI using specialist agents
Governance	Quality gate + risk engine
Best For	Solo developers, minimal overhead

`--profile team` (5-15 people)

npx atabey init gemini --profile team --unified

Feature	What You Get
Agents	5-8 focus-specific agents (manager, architect, backend, quality, database, security)
Setup Time	~30 seconds
Dashboard	Extensive live modules (Agent Monitor, Hermes, Approvals, Compliance, FinOps, Loop Detector, Memory, etc.) with WebSocket real-time updates
Governance	Quality gate + risk engine + contract validation
Best For	Small-medium teams with governance needs

`--profile enterprise` (15+ people)

npx atabey init gemini --profile enterprise

Feature	What You Get
Agents	All 13 agents (Supreme + Core + Recon)
Setup Time	~1 minute
Security	Human-in-the-Loop, KVKK PII masking, audit log
Governance	Full governance with circuit breakers
Best For	Enterprise with compliance requirements

13 Specialized Agents

[!NOTE] Execution Model: Atabey's specialized agents are virtual profiles powered by dynamically injected system role definitions and context templates. They do not run as 13 parallel, independent LLM API instances. Instead, your single host AI assistant (e.g., Claude Code, Gemini CLI) takes on these roles sequentially, routed by the TF-IDF engine, while Atabey enforces boundaries, state machine constraints, and quality checks during role switches.

Agent	Tier	Capability	Role	Freelancer	Team	Enterprise
@manager	Supreme	10/10	Orchestration, governance, quality gate	✅	✅	✅
@security	Supreme	10/10	Security audit, vulnerability scanning	❌	✅	✅
@architect	Core	9/10	System design, contracts, architecture	❌	✅	✅
@backend	Core	9/10	Backend dev, API, business logic, tests	✅	✅	✅
@frontend	Core	9/10	UI, atomic components, responsive design	❌	✅	✅
@quality	Core	9/10	Compliance, lint, test coverage	✅	✅	✅
@database	Core	9/10	Database management, migrations, queries	❌	✅	✅
@analyst	Recon	8/10	Strategy analysis, requirements	❌	❌	✅
@mobile	Core	8/10	React Native mobile development	❌	❌	✅
@native	Recon	8/10	Native platform integration	❌	❌	✅
@devops	Core	8/10	CI/CD, deploy, infrastructure	❌	❌	✅
@explorer	Recon	8/10	Codebase discovery, analysis	❌	❌	✅
@git	Recon	8/10	Version control, commit management	❌	❌	✅
@human	—	—	Human approval placeholder	✅	✅	✅

Agent Hierarchy

User / External Request
    ↓
  @manager (Supreme Authority)
    ↓
  @security (Parallel — always watching)
    ↓
  @architect → @backend / @frontend / @database / @devops
    ↓
  @quality → @mobile / @native / @explorer / @git / @analyst

39 MCP Tools

File System (6)

read_file, view_file, write_file, replace_text, batch_surgical_edit, patch_file

Search & Exploration (4)

list_dir, grep_search, get_project_map, get_project_gaps

Framework & System (10)

run_shell_command, run_tests, get_system_health, check_active_ports, get_framework_status, read_project_memory, get_memory_insights, update_project_memory, orchestrate_loop, submit_plan

Control Plane (3)

acquire_lock, release_lock, register_agent

Memory & Knowledge (4)

store_knowledge, search_knowledge, delete_knowledge, update_contract_hash

Messaging & Human (3)

send_agent_message, log_agent_action, ask_human

Compliance & Quality (4)

mask_pii, analyze_code_quality, check_architecture_compliance, check_lint

Operations (3)

approve_operation, compress_files, decompress_files, http_proxy_request, audit_dependencies

7 Core Skills

Skill	Tools	Description
FILE_SYSTEM	read_file, write_file	Token-efficient file operations
EDITING	replace_text, patch_file	Surgical code modification
ORCHESTRATION	orchestrate_loop, send_agent_message, get_framework_status, log_agent_action	Hermes messaging
GOVERNANCE	acquire_lock, release_lock, register_agent, update_contract_hash	Resource locking & contracts
QUALITY_ASSURANCE	run_shell_command, view_file	Testing & code quality
DATABASE_MANAGEMENT	view_file, replace_text, run_shell_command	DB & migrations
DEVOPS_INFRASTRUCTURE	run_shell_command, view_file	CI/CD & deployment

Each skill adapts to the target platform (Claude → native tool names, Gemini → YAML, Codex → categories).

3-Layer Memory System

Layer	Storage	Technology	Purpose
Vector Memory	SQLite	TF-IDF / OpenAI embeddings + Cosine similarity	Semantic search across project knowledge
Project Memory	Markdown	`PROJECT_MEMORY.md`	Central project state, synchronized every session
Specialty Memory	Markdown	`.atabey/memory/specialties/{agent}.md`	Per-agent learned conventions from past tasks

Memory Flow:

Task → Evaluation Engine → Score (0-100)
  ├── Success → extractSuccessLesson() → updateSpecialtyMemory()
  └── Failure → compliance/lint/test lesson → updateSpecialtyMemory()
       ↓
  Next session: readLearnedConventions() → injected into agent prompt

Knowledge Base (30+ Standards)

Category	Standards
Governance	governance-standards, llm-governance, crud-governance
Architecture	architecture-standards, auth-standards
Development	frontend-standards, nextjs-standards, mobile-standards, kysely-standards
Security	security-standards, security-audit-standards, logging-and-secrets
DevOps	deployment-standards, github-actions-standards
Quality	quality-standards, testing-standards, vitest-standards, playwright-standards
Performance	performance-standards, observability-standards
UI/UX	tailwind-standards, react-query-standards, react-router-standards
Data	typeorm-standards, vite-standards
Economics	token-economy, pino-standards, swagger-standards

Agent knowledge files are automatically embedded into system prompts at build time.

Multi-Layer Governance Pipeline

Every MCP tool call passes through this pipeline:

CallToolRequest
    ↓
 1. [VALIDATION]      Zod schema validation
 2. [GOVERNANCE]      validateArgsAgainstRules()
 3. [DISCIPLINE]      enforceDiscipline()
 4. [LOOP DETECTION]  recordAndCheck() + cooldown
 5. [FINOPS]          budgetManager.recordUsage()
 6. [LICENSE]         validateLicenseCompliance()
 7. [AUTO-ROLLBACK]   prepareWrite() snapshot
 8. [HUMAN-IN-LOOP]   RiskEngine.assessTaskRisk() + checkRiskGate()
 9. [PII MASKING]     maskToolArgs() → mask before handler
 10. [EXECUTION]      Tool handler runs
 11. [POST-VALIDATION] scanFileForViolations() → rollback if needed
 12. [INJECTION PROTECTION] sanitizeResponse()
 13. [PII MASKING]    maskToolResult() → mask before returning to AI
    ↓
  Return governed result

Core Features

Deterministic Quality Gate

No agent can push code directly to production. All outputs pass through AST analysis (compliance) + linting + governance validation.

Smart Routing Engine

TF-IDF + Semantic (cosine similarity) routing with 60/40 blend. Falls back to keyword-based routing when embeddings are unavailable.

Hermes Message Broker

SQLite-backed async message queue for inter-agent communication. Lock-based protocol prevents race conditions.

Risk Engine (Human-in-the-Loop)

Operations containing DROP, DELETE, TRUNCATE, or secret manipulation are flagged. Score ≥ 60 requires human approval.

Specialty Memory (Agent Learning)

Agents learn from both successes and failures. Lessons stored in .atabey/memory/specialties/ and auto-injected next session.

Auto-Rollback

Pre-write snapshots, post-write governance scan, automatic rollback + regenerate instructions for the AI.

Token Economy & Cost Tracking

Per-agent token/cost tracking via MCP governance middleware. Budget enforcement is config-driven in .atabey/config.json (finops section) and auto-starts with the MCP server. See Token Economy & FinOps.

Prompt Injection Protection

OWASP LLM01 compliant sanitization detected via role-playing, system prompt override, delimiters, and context manipulation patterns.

Multi-Client MCP (Stdio + HTTP/SSE)

Two transport modes: Stdio (per-developer) and HTTP/SSE (shared server for entire team, port 5858).

Dashboard

npx atabey dashboard  # Opens at http://localhost:5858

Module	Description	Update
🤖 Agent Monitor	AI agent status + tasks	WS (5s)
📨 Hermes Stats	Message queue statistics	WS (5s)
💬 Hermes Messages	Message queue viewer + filter	WS (5s)
🔐 Approval Center	Human-in-the-Loop approvals	WS
📋 Task Planner	Task DAG + progress	REST (5s)
📝 Agent Logs	Execution logs	WS (5s)
⚠️ Error Tracker	Lint/compliance/security errors	WS
🧠 Memory Insights	Vector memory search	REST
🛡️ Compliance	Quality gate violations	REST (15s)
✅ Quality Panel	Code quality analysis	REST
🔌 Adapters	Adapter-skill mapping	REST
🔒 GDPR / KVKK	PII masking, erasure, audit trail	REST

CLI Reference

atabey init [adapter]           Initialize Atabey (--profile freelancer|team|enterprise)
atabey mcp start|install        MCP server management
atabey dashboard [port]         Open web dashboard (default: 5858)
atabey orchestrate              Start orchestration loop
atabey check                    System health check
atabey status                   Show agent statuses and costs
atabey plan [submit <json>]     Task planning
atabey approve <traceId>        Approve blocked high-risk task
atabey trace new|replay <id>    Execution traces
atabey memory update <sect> <c> Update project memory
atabey git commit|sync          Git operations
atabey explorer graph|audit     Code analysis
atabey security <path>          Security audit
atabey compliance <path>        Run compliance check
atabey contract                 Verify contract integrity
atabey lint                     Run ESLint
atabey knowledge <query>        Knowledge base search
atabey log <agent> <msg>        View agent logs
atabey script <name> [dir]      Run predefined script
atabey index [dir]              Index codebase for RAG
atabey coverage                 Test coverage reports
atabey quickstart               Generate example task file

`atabey init [adapter]` Options

Option	Values	Description
`[adapter]`	`gemini`, `claude`, `cursor`, `grok`, `codex`, `local`, `antigravity-cli`	Target AI platform
`--profile`	`freelancer`, `team`, `enterprise`	Preset agent group layout
`--focus`	`fullstack`, `backend`, `frontend`, `mobile`, `mobile-fullstack`	Project type optimization
`--lang`	`tr`, `en`	Constitution language
`--unified`	Flag	Multi-platform export
`--yes`	Flag	Non-interactive mode

Architecture

┌──────────────────────────────────────────────────────────────────┐
│                      AI Chat Interface                            │
│    @backend Create login service (in Claude/Gemini/Cursor)      │
├──────────────────────────────────────────────────────────────────┤
│                    MCP Server (atabey-mcp/)                       │
│                    39 Tools · Multi-Layer Governance                │
│                    Zod Validation · PII Masking                  │
├──────────────────────────────────────────────────────────────────┤
│                    src/cli/  30+ Commands                         │
│                    src/modules/  Engines + Agents + Memory       │
│                    src/shared/  Types + Storage + Audit          │
├────────────────┬───────────────────────────┬──────────────────────┤
│                ▼                           ▼                      │
│  ┌──────────────────────────┐  ┌──────────────────────────────┐  │
│  │ Web Dashboard            │  │ SQLite (better-sqlite3)      │  │
│  │ Port 5858                │  │ Vector Memory · Hermes Queue │  │
│  │ 12 Module WS Live        │  │ Audit Logs · Locks · Tasks  │  │
│  │ Responsive (Mob+Desk)    │  │ Knowledge · Agent Registry  │  │
│  └──────────────────────────┘  └──────────────────────────────┘  │
└──────────────────────────────────────────────────────────────────┘

Security

Deterministic & Rule-Based Governance

Deterministic rules: AST compliance parsing, strict TypeScript type validation (zero any), and automated governance checks.

Zero Type Hole Policy

any type usage is strictly forbidden
All function inputs validated with Zod schemas
Type safety enforced in governance pipeline

PII Masking (KVKK/GDPR Technical Alignment)

All logs scanned for Personally Identifiable Information
Sensitive data automatically masked (20+ patterns)
Designed to assist with Turkish KVKK (Law No. 6698) and EU GDPR controls
Right to Erasure (KVKK Art. 7 / GDPR Art. 17) technical support

Human-in-the-Loop

Risk score ≥ 60 requires human approval
In-chat approval: approve_operation MCP tool
Terminal fallback: atabey approve <traceId>

AI Discipline Engine

Rate Limiting: Max 60 calls/minute per agent
File Size Limits: Prevents >1MB files
Loop Detection: Blocks >10 consecutive same-tool calls
Cooldown Mechanism: Automatic when limits exceeded
Injection Protection: Aligned with OWASP LLM01 guidelines

KVKK/GDPR Technical Alignment

Scope: Technical compliance tooling — not legal certification. Pair with your DPO/legal counsel for production deployments.

Feature	KVKK	GDPR	Status
PII Masking (20+ patterns)	Art. 4, 5, 11, 12	Art. 5, 32	✅ Implemented
Data Retention (30/90 days)	Art. 5, 7	Art. 5, 17	✅ Auto-cleanup on MCP boot
Right to Erasure	Art. 7	Art. 17	✅ `/api/audit/erase` + trace erase
Audit Trail	Art. 11	Art. 30	✅ SQLite audit log
Data Classification	Art. 6	Art. 9	✅ Category-based retention
Consent Logging	Art. 5	Art. 7	✅ `.atabey/compliance/consent-log.json`

Configure in .atabey/config.json:

{
  "compliance": {
    "retentionEnabled": true,
    "consentLogging": true,
    "piiMasking": true,
    "dataProcessingBasis": "consent"
  }
}

Dashboard API: GET /api/compliance/consent, GET /api/compliance/retention

EU AI Act Alignment

Atabey provides process and governance alignment with the EU AI Act via .atabey/knowledge/llm-governance.md (scaffolded on atabey init):

EU AI Act principle	Atabey implementation
Human oversight	Human-in-the-Loop, risk score ≥ 60 → approval
Risk classification	High-risk ops (auth, billing, moderation) gated by `@manager`
Transparency	AI-generated output labeling rules in llm-governance
Data protection	PII masking, retention, erasure (KVKK/GDPR layer)
Robustness & safety	Loop detection, injection protection (OWASP LLM01), auto-rollback

Not included: CE marking, conformity assessment, DPIA templates, or legal EU AI Act certification. Use Atabey as a technical control layer alongside legal review.

Token Economy & FinOps

FinOps auto-starts when the MCP server boots. Defaults are profile-based:

Profile	Tracking	Budget enforcement
`freelancer`	✅ Always	❌ Off (track only)
`team`	✅ Always	❌ Off by default
`enterprise`	✅ Always	✅ $500/mo default

Configure in .atabey/config.json:

{
  "finops": {
    "tracking": true,
    "enforcement": true,
    "monthlyBudgetUsd": 200,
    "agentMaxBudgetUsd": 40,
    "team": "my-team",
    "alertThresholds": [50, 80, 90, 100],
    "costPer1kTokensUsd": 0.003
  }
}

Environment overrides (optional):

ATABEY_BUDGET_ENABLED=true
ATABEY_BUDGET_MONTHLY=200
ATABEY_BUDGET_AGENT_MAX=40

Tracking mode: logs estimated tokens per MCP tool call, exposes /api/finops
Enforcement mode: blocks tool calls when monthly/agent budget exceeded
Alerts: stderr warnings at 50/80/90/100% thresholds

Check status: atabey status · Dashboard: GET /api/finops

Testing

npm test                    # Run all tests
npm run test:watch          # Watch mode
npm run test:coverage       # Coverage report

Contributing

Please read CONTRIBUTING.md for details on the code of conduct and pull request process.

git clone https://github.com/ysf-bkr/atabey.git
cd atabey
npm install
npm run build

License & Business Model

Code: GNU Affero General Public License v3.0 — Yusuf BEKAR

This program is free software: you can redistribute it and/or modify it under the terms of the GNU Affero General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

Network Use Clause (Section 13): If you modify the Program and make it accessible over a network (e.g., as a SaaS service), you must provide the complete corresponding source code to all users who interact with it remotely.

Service Model:

Enterprise Support & SLA — Guaranteed response times, priority bug fixes, custom integrations
Consulting & Training — Team onboarding, governance policy design, architecture review
Managed Enterprise Server — Centralized telemetry, multi-team budget management, org-wide dashboard

Enterprise inquiries: [email protected]

Developer: Yusuf BEKAR — "Order from Chaos"

Türkçe — Agent Atabey

Dil notu: Proje kodu, şablonları, standartları, CLI metinleri ve ajan tanımları İngilizcedir. Yalnızca bu kök README.md hem İngilizce hem Türkçe sunulur. Paket README dosyaları (packages/*/README.md) İngilizcedir.

Agent Atabey nedir?

Agent Atabey, MCP (Model Context Protocol) üzerine kurulu bir AI Yönetişim ve Çok-Ajan Platformu / Orkestratörüdür. Claude Code, Gemini CLI, Cursor, Codex, Grok, Antigravity veya yerel LLM gibi mevcut AI kodlama arayüzlerinizin üzerine oturur; onları yönetişimli, çok-ajanlı bir yazılım mühendisliği sistemine dönüştürür.

Katman	Rol
AI IDE/CLI’niz	Kod yazar, tool çalıştırır, shell komutları yürütür
Atabey MCP Server	~39 tool · çok katmanlı yönetişim · orkestrasyon · bellek
@manager + 13 uzman	Delegasyon, kalite kapısı, risk kontrolü, denetim izi
`.atabey/` beyin merkezi	Anayasa, bellek, knowledge, registry, gözlemlenebilirlik

[!NOTE] Proje durumu (Pre-1.0 / açık kaynak deneyi): Atabey şu an pre-alpha (v0.0.x) aşamasındadır ve tek geliştirici tarafından aktif geliştirilmektedir. Çekirdek özellikler (TF-IDF routing, regex PII maskeleme, deterministik kurallar) çalışır; kurumsal üretim ortamları için henüz hazır değildir.

Kullanmadan önce bilinen sınırlar için BLINDSPOTS.md dosyasını okuyun.

Felsefe: “Kaostan Nizam” / Order from Chaos

Atabey nedir: AI kodlama asistanları için deterministik, kural tabanlı yönetişim ve çok-ajan orkestrasyon middleware’i.

Atabey ne değildir: Bağımsız bir LLM runtime (LangGraph/CrewAI gibi) veya ML tabanlı tehdit analiz platformu değildir. Kodu bağlı AI asistanı üretir; Atabey yönlendirir, disipline eder, doğrular ve koordine eder.

Nasıl çalışır?

Atabey, AI arayüzünüze MCP tool sunucusu olarak bağlanır. Bağlandıktan sonra sohbette @agent komutlarını kullanırsınız:

Siz (@backend): "JWT kimlik doğrulamalı kullanıcı login servisi yaz"
     │
     ▼
  Atabey MCP Server (atabey-mcp/)
     │  @agent komutunu yakalar
     │  RoutingEngine ile yönlendirir (TF-IDF + semantik)
     │  @backend uzmanına yollar
     │  Yönetişim pipeline doğrular
     │  Quality Gate çıktıyı kontrol eder
     │  Vector Memory’e kaydeder
     │
     ▼
  Yönetilmiş, denetlenmiş, izli kod döner

Günlük kullanımda ayrı bir terminal gerekmez. AI sohbetinde @agent sözdizimi yeterlidir.

Önemli: Atabey kod yazmaz. Kodu Claude Code / Gemini / Cursor gibi bağlı asistan yazar. Atabey:

Routing — doğru uzman profile (@backend, @security, …)
Risk Engine — tehlikeli işlemleri skorlar, gerekirse insan onayı ister
Quality Gate — standartlara, sözdizimine, test metriklerine göre denetler
Memory — mimari kararlar ve uzmanlık geçmişini tutar

Platform mimarisi

Kullanıcı / Geliştirici
  → Claude Code / Gemini CLI / Cursor IDE
  → MCP (stdio)
  → Atabey MCP Server
       Yönetişim · Hermes kuyruk · Orchestrator · Vector Memory
       Risk · FinOps · Auto-Rollback · Human-in-the-Loop
  → .atabey/  (beyin)  ·  .agents/  (birleşik ajan hub)

Monorepo paketleri:

Paket	Amaç
`packages/atabey`	CLI, adapter’lar, ajan export, init
`packages/atabey-mcp`	MCP sunucu, yönetişim middleware, dashboard API
`packages/shared`	Ortak constants, PII, audit, storage (SSOT)

Hızlı başlangıç

# Tek platform
npx atabey init gemini --profile freelancer --yes

# 7 platform (ekip için önerilir)
npx atabey init gemini --unified --profile team --yes

# Sağlık kontrolü
npx atabey check
npx atabey status

MCP bağlama (özet — Claude / Cursor):

{
  "mcpServers": {
    "atabey": {
      "command": "npx",
      "args": ["-y", "atabey-mcp"],
      "env": {
        "MCP_TRANSPORT": "stdio",
        "ATABEY_PROJECT_ROOT": "/path/to/your/project",
        "ATABEY_AUTO_START_ORCHESTRATOR": "true"
      }
    }
  }
}

Sohbette:

@backend Create a REST API for user management with CRUD operations
@security Audit the authentication middleware
@quality Run compliance check on the new feature

İçerik dili: Oluşturulan anayasa, standartlar, ajan talimatları ve CLI metinleri İngilizcedir. --lang bayrağı içerik dilini değiştirmez (geriye dönük uyumluluk için kabul edilebilir).

5 temel yetenek

#	Yetenek	Açıklama
🛠️	~39 MCP Tool	Dosya sistemi, arama, mesajlaşma, yönetişim, bellek, kalite, ağ, orkestrasyon
🧠	3 katmanlı bellek	Vector (TF-IDF/OpenAI), Project Memory, Specialty Memory
🤖	13 uzman ajan	Supreme / Core / Recon hiyerarşisi
🎯	7 core skill	7 AI platformuna uyarlanan skill setleri
📚	30+ knowledge standard	Yönetişim, güvenlik, mimari, test, deployment

Desteklenen platformlar

Platform	MCP	Not
Claude Code	`.mcp.json`	En güçlü orkestrasyon
Gemini CLI	`.gemini/mcp.json`	Güçlü komutan / stratejist
Cursor	`.cursor/mcp.json`	Implementer / editör
Codex / Copilot	`.vscode/mcp.json`	Sınırlı skill seti
Antigravity	`.agents/mcp_config.json`	Custom agent JSON
Grok	`.grok/mcp_config.json`	Sınırlı skill seti
Local LLM	`.atabey/mcp_config.json`	Offline / private

npx atabey init claude --unified   # tüm platformlara export

Kurulum gereksinimleri

Gereksinim	Sürüm
Node.js	>= 18
npm	>= 9
AI arayüzü	Claude, Gemini, Cursor, Codex, Grok, Antigravity veya local LLM

npm install -g atabey
npx atabey init gemini --profile freelancer --yes
npx atabey status
npx atabey dashboard   # opsiyonel web UI (port 5858)

better-sqlite3 native bağımlılıktır; CI’da derleme sorunlarında npm install --ignore-scripts ve önceden derlenmiş binary’ler gerekebilir.

Profiller

Profil	Kim için	Ne gelir
`--profile freelancer`	1–3 kişi	Az ajan, düşük sürtünme
`--profile team`	5–15 kişi	5–8 odak ajanı, unified önerilir
`--profile enterprise`	15+	Tam 13 ajan, sıkı yönetişim

Odak: --focus fullstack|backend|frontend|mobile

13 uzman ajan (özet)

Ajan	Katman	Rol
`@manager`	Supreme	Orkestrasyon, onay zinciri
`@security`	Supreme	Güvenlik denetimi
`@architect`	Core	Tasarım, kontratlar
`@backend` / `@frontend`	Core	API / UI
`@quality`	Core	Uyumluluk, lint, coverage
`@database` / `@devops`	Core	Veri / CI-CD
`@analyst` / `@mobile` / `@native`	Core	Strateji / mobil / native
`@explorer` / `@git`	Recon	Keşif / versiyon kontrol

MCP tool grupları

Dosya sistemi: read/write, replace, patch, batch edit
Shell: allow-list’li run_shell_command (+ opsiyonel uid/gid sandbox)
Arama: list_dir, grep, project map/gaps
Mesajlaşma: send_message, ask_human, approve, log
Bellek / knowledge: store/search/delete, insights
Framework: status, orchestrate, plan, tests, audit_deps
Kalite / compliance / control plane / observability / network

3 katmanlı bellek

Vector Memory — TF-IDF veya OpenAI embedding + cosine
Project Memory — PROJECT_MEMORY.md
Specialty Memory — ajan bazlı öğrenilmiş dersler

Yönetişim pipeline (özet)

Her tool çağrısında (pre/post): PII mask → schema validation → rules → discipline → silent routing → CRUD governance → loop detection → FinOps → license → auto-rollback snapshot → risk/HITL → execute → response checks → injection sanitize → audit / WebSocket.

Yerel, harici bağımlılıksız ek katmanlar:

In-memory Job Queue — eşzamanlı ajan işi sınırı
Worker threads pool — CPU işleri
Atomic file lock (wx) — dosya yarışları
uid/gid sandbox — shell izolasyonu (yapılandırıldığında)
Circuit breaker + exponential backoff — geçici hata dayanıklılığı

Dashboard

npx atabey dashboard
# veya
MCP_TRANSPORT=unified MCP_PORT=5858 npx atabey-mcp

Paneller: Agents, Hermes, HITL Approvals, Plans, Logs, Quality, FinOps, Loop Detector, License, Telemetry, Memory, Compliance, …

CLI (seçme)

Komut	Açıklama
`atabey init …`	Proje iskeleti + adapter export
`atabey check` / `status`	Sağlık / ajan durumu
`atabey mcp start\|install`	MCP yönetimi
`atabey orchestrate`	Etkileşimli orkestrasyon döngüsü
`atabey plan` / `approve` / `trace`	Plan, onay, iz
`atabey memory` / `knowledge`	Bellek / bilgi bankası
`atabey dashboard`	Web UI

Güvenlik ve uyumluluk (dürüst özet)

Allow-list shell, risk skoru, HITL, PII maskeleme, loop detection
Sandbox varsayılan kapalıdır — ATABEY_SANDBOX_UID / USER ile açılır
KVKK / GDPR / EU AI Act: teknik hizalama kontrolleri vardır; yasal “tam uyumluluk” iddiası yoktur
Detay: SECURITY.md, PRIVACY.md, BLINDSPOTS.md

FinOps

Token tahmini kabaca metin uzunluğu / 4’tür; gerçek tokenizer faturalaması değildir. Bütçe uyarıları config.json / env ile yapılandırılır.

Test ve katkı

npm test
npm run test:watch
npm run test:coverage

Katkı süreci: CONTRIBUTING.md

git clone https://github.com/ysf-bkr/atabey.git
cd atabey && npm install && npm run build

Lisans ve iş modeli

Kod: AGPL-3.0 — Yusuf BEKAR

Ağ üzerinden (SaaS) sunulursa kaynak kod paylaşımı gerekir (AGPL §13).

Hizmet: kurumsal destek/SLA, danışmanlık/eğitim, yönetilen enterprise sunucu. İletişim: [email protected]

Geliştirici: Yusuf BEKAR — “Order from Chaos”

from github.com/ysf-bkr/atabey

Установить Atabey в Claude Desktop, Claude Code, Cursor

Рекомендуется · одна команда, все IDE

unyly install atabey

Ставит в Claude Desktop, Claude Code, Cursor и VS Code — сам разбирается с npx, uvx и сборкой из исходников.

Впервые? Поставь CLI: curl -fsSL https://unyly.org/install | sh

Или настроить вручную

Выполни в терминале:

claude mcp add atabey -- npx -y atabey

FAQ

Atabey MCP бесплатный?

Да, Atabey MCP бесплатный — установка в пару кликов через Unyly без оплаты.

Нужен ли API-ключ для Atabey?

Нет, Atabey работает без API-ключей и переменных окружения.

Atabey — hosted или self-hosted?

Self-hosted: сервер запускается локально на твоей машине командой из раздела установки.

Как установить Atabey в Claude Desktop, Claude Code или Cursor?

Открой Atabey на unyly.org, выбери вкладку своего клиента (Claude Desktop, Claude Code, Cursor) и нажми Install — конфиг сгенерируется автоматически, без правки JSON.

Compare Atabey with

AtabeyvsGitHub AtabeyvsFilesystem AtabeyvsMemory AtabeyvsTemplate MCP Server

Не уверен что выбрать?

Найди свой стек за 60 секунд

Автор?

Embed-бейдж для README

Похожее

Все в категории development

Command Palette

Atabey

Описание

README

🚀 How It Works

🧠 How Atabey Relates to Your AI

📋 Table of Contents

Platform Architecture

Orchestration Flow

Project kinds

Quick Start

1. Initialize Atabey in Your Project

2. Verify MCP & Platform Health

3. Connect to Your AI Interface (MCP auto-starts)

MCP Connection (IDE/CLI)

4. Start Using in AI Chat

5 Core Capabilities Overview

How Atabey Plugs Into Your AI

Supported Platforms

Installation

Requirements

Quick Install

Native Dependency & CI/CD Considerations

Profile-Based Setup

--profile freelancer (1-3 people)

--profile team (5-15 people)

--profile enterprise (15+ people)

13 Specialized Agents

Agent Hierarchy

39 MCP Tools

File System (6)

Search & Exploration (4)

Framework & System (10)

Control Plane (3)

Memory & Knowledge (4)

Messaging & Human (3)

Compliance & Quality (4)

Operations (3)

7 Core Skills

3-Layer Memory System

Knowledge Base (30+ Standards)

Multi-Layer Governance Pipeline

Core Features

Deterministic Quality Gate

Smart Routing Engine

Hermes Message Broker

Risk Engine (Human-in-the-Loop)

Specialty Memory (Agent Learning)

Auto-Rollback

Token Economy & Cost Tracking

Prompt Injection Protection

Multi-Client MCP (Stdio + HTTP/SSE)

Dashboard

CLI Reference

atabey init [adapter] Options

Architecture

Security

Deterministic & Rule-Based Governance

Zero Type Hole Policy

PII Masking (KVKK/GDPR Technical Alignment)

Human-in-the-Loop

AI Discipline Engine

KVKK/GDPR Technical Alignment

EU AI Act Alignment

Token Economy & FinOps

Testing

Contributing

License & Business Model

Türkçe — Agent Atabey

Agent Atabey nedir?

Nasıl çalışır?

Platform mimarisi

Hızlı başlangıç

5 temel yetenek

Desteklenen platformlar

Kurulum gereksinimleri

Profiller

13 uzman ajan (özet)

MCP tool grupları

3 katmanlı bellek

`--profile freelancer` (1-3 people)

`--profile team` (5-15 people)

`--profile enterprise` (15+ people)

`atabey init [adapter]` Options