A Rust-based self-hosted multi-tenant AI Agent platform with trait-driven modular architecture, supporting 15 LLM providers, 9 communication channels, and MCP protocol extension, deployable from Raspberry Pi to cloud servers, suitable for Vietnam market and enterprises requiring data localization.
Core Capabilities#
LLM Providers (15)#
OpenAI, Anthropic, Gemini, DeepSeek, Groq, OpenRouter, Together, MiniMax, xAI (Grok), Mistral, Ollama, llama.cpp, Brain Engine (local inference), CLIProxy, vLLM + custom endpoint
Communication Channels (9)#
CLI, Telegram, Discord, Email (IMAP/SMTP), Webhook, WhatsApp, Zalo (Personal + Official Account)
Built-in Tools (13 + MCP Extension)#
- File Operations: Shell, File, Edit File, Glob, Grep
- Network: Web Search, HTTP Request
- Code Execution: Execute Code (9 languages)
- Management: Config Manager, Plan Mode, Group Summarizer, Calendar
- Knowledge: Document Reader, Memory Search, Session Context
- MCP Bridge: Connect to any MCP Server (GitHub, Postgres, Brave Search, etc.)
Platform Features#
| Feature | Description |
|---|---|
| Multi-Tenant | Admin Dashboard, JWT Auth, Tenant Manager, Pairing Codes, Audit Log, Per-tenant SQLite DB |
| Web Dashboard | 12-page UI, bilingual (Vietnamese/English), WebSocket real-time communication |
| Agent Gallery | 51 business Agent templates across 13 business categories |
| Group Chat | Multi-Agent collaboration, single query triggers parallel specialist Agent responses |
| 3-Tier Memory | Brain workspace (SOUL.md/MEMORY.md), Daily auto-compaction, FTS5 search |
| Knowledge RAG | Document upload → vector search → relevance scoring |
| Scheduler | Scheduled tasks, Agent background auto-execution |
| Brain Engine | GGUF inference: mmap, quantization, Flash Attention, SIMD (ARM NEON, x86 SSE2/AVX2) |
| Security | Command allowlist, AES-256, HMAC-SHA256, JWT + bcrypt, CORS, rate limiting |
Architecture#
Crate Structure (14)#
- bizclaw-core: Traits, types, config, errors
- bizclaw-brain: GGUF inference + SIMD (ARM NEON, x86 AVX2)
- bizclaw-providers: 15 LLM providers (OpenAI-compatible unified interface)
- bizclaw-channels: 9 channel types
- bizclaw-memory: SQLite + FTS5, Brain workspace, daily auto-compaction
- bizclaw-tools: 13 native tools + MCP bridge
- bizclaw-mcp: MCP client (JSON-RPC 2.0 via stdio)
- bizclaw-security: AES-256, Command allowlist, Sandbox
- bizclaw-agent: Agent loop, tool calling (max 3 rounds), context management
- bizclaw-gateway: Axum HTTP + WS + Dashboard (12 pages, i18n VI/EN)
- bizclaw-knowledge: Knowledge RAG with FTS5, document chunking
- bizclaw-scheduler: Scheduled tasks, agent integration, notifications
- bizclaw-runtime: Process adapters
- bizclaw-platform: Multi-tenant admin platform, JWT, audit log
Tech Stack#
- Runtime: Tokio (async)
- HTTP/WebSocket: Axum
- Database: SQLite + FTS5
- Inference: GGUF + SIMD (Flash Attention, FP16 KV Cache)
- Security: JWT + bcrypt, AES-256, HMAC-SHA256
- Containerization: Docker + docker-compose
Deployment#
One-click Install (VPS/Pi)#
curl -sSL https://bizclaw.vn/install.sh | sudo bash
Docker#
git clone https://github.com/nguyenduchoai/bizclaw
cd bizclaw && docker-compose up -d
Build from Source#
git clone https://github.com/nguyenduchoai/bizclaw.git
cd bizclaw && cargo build --release
./target/release/bizclaw-platform --port 3001
Deployment Modes#
| Mode | Binary | Use Case |
|---|---|---|
| Standalone | bizclaw only | Single bot, personal, testing |
| Platform | bizclaw + bizclaw-platform | Multi-bots, agency, production |
Requirements#
- Minimum: Raspberry Pi 512MB RAM
- Supported: Linux, macOS, Windows
- Binary size: bizclaw ~12MB, platform ~7.7MB
- Version: v0.2.0
Business Scenarios#
Enterprise Coverage (13 Categories)#
- HR: Recruitment, onboarding, compensation, KPI, policies
- Sales: CRM, quotations, sales, telesales, partnerships
- Finance: Accounting, tax, cash flow, invoicing, internal controls
- Operations: Warehouse, procurement, logistics, QC, maintenance
- Legal: Contracts, compliance, IP, disputes
- Customer Service: Support, tickets, CSAT & feedback
- Marketing: Content, SEO, ads, social media, branding
- IT: Helpdesk, cybersecurity, infrastructure
Cost Optimization#
Each Agent can independently select Provider, saving 60-80% API costs compared to unified high-price Provider usage