Langboard

Langboard is an AI Agent Orchestration Platform that helps organizations run AI workflows with operational control and human oversight.

Unlike traditional automation tools or fully autonomous agent experiments (e.g., AutoGPT, BabyAGI), Langboard strikes the balance between AI autonomy and human governance, ensuring safe, scalable, and enterprise-ready AI orchestration.

---

🌍 Vision and Purpose

AI systems are powerful, but not infallible. Even state-of-the-art large language models (LLMs) hallucinate, misinterpret context, or act unpredictably. Langboard was created with a simple principle:

"AI should act, but humans must approve."

Langboard's mission is to enable enterprises to harness AI efficiency without sacrificing reliability, compliance, or accountability.

---

🚀 Quick Start

• Prerequisite: Install Docker and Docker Compose first.

• For Windows users

  • You can run one of quickstart.ps1, quickstart-ollama-cpu.ps1, and quickstart-ollama-gpu.ps1 in scripts directory with powershell
    • If the system opens Notepad when you double-click the files, you should select 'Run with powershell' by right-click menu.
  • You can run powershell scripts below

  cd .\scripts
  .\quickstart.ps1    REM default
  .\quickstart-ollama-cpu.ps1   REM if you want CPU mode
  .\quickstart-ollama-gpu.ps1   REM if you want GPU mode

• For other OS users,

  • You can double-click one of quickstart.sh, quickstart-ollama-cpu.sh, and quickstart-ollama-gpu.sh in scripts directory
  • You can run bash scripts below

  cd ./scripts
  ./quickstart.sh              # default
  ./quickstart-ollama-cpu.sh   # if you want CPU mode
  ./quickstart-ollama-gpu.sh   # if you want GPU mode

---

✨ Platform Overview

1. Kanban-native workspace

   • Manage work through boards, columns, and cards.
   • Built-in checklists, comments, attachments, labels, relationships, and wiki pages.

2. Real-time collaboration

   • Socket-based updates across boards, cards, wiki, notifications, and bot events.
   • Live editor session events and streaming responses for AI interactions.

3. AI-assisted work execution

   • Project bots and internal bots can run tasks, schedules, and scoped automations.
   • Editor chat/copilot graph execution is available for card and wiki writing workflows.

---

🤖 AI Orchestration with LangGraph

• Supported bot platforms: default, Langflow, n8n.
• The built-in default platform runs on LangGraph.
• External Langflow remains supported through user-provided endpoint bots.
• Graph execution endpoints: /api/v1/graph/run/{bot_id}, /api/v1/graph/resume/{thread_id}, and /api/v1/graph/status/{thread_id}.
• Built-in graph execution supports human-in-the-loop interrupt/resume, bot triggers, schedules, freeze state, and scoped project/card context.

---

🧩 MCP Integration

• Embedded FastMCP server is mounted at /mcp (streamable HTTP transport).
• MCP tool groups support admin/global and user-scoped governance.
• Requests must include X-MCP-Tool-Group-UID for tool-group validation.
• Middleware enforces authentication, ownership checks, and MCP role permissions.
• Built-in MCP tools cover project, card, bot, activity, and metadata operations.

---

🔐 API Keys and Key Vault

• API key lifecycle operations include create, update, activate/deactivate, expiration, and delete.
• IP whitelist validation and API key usage logging are built in.
• Key material is issued through KeyVault providers.
• Supported providers: OpenBao, HashiCorp Vault, AWS KMS, Azure Key Vault.

---

🛡️ Governance and Access Control

• Role-based controls are applied to settings, API keys, MCP, bots, and user management.
• Route-level authorization is enforced through auth and role filters.
• Central settings APIs cover users, bots/internal bots, webhooks, API keys, MCP tool groups, and global relationships.

---

🏗️ Architecture

Langboard combines five operational layers:

• Workspace layer: board-centric collaboration and AI-assisted editing UI.
• Application layer: API and Socket services for orchestration and real-time events.
• Graph layer: LangGraph runtime service for internal graph execution, checkpoints, and resumable human-in-the-loop events.
• Tool layer: MCP gateway with role-aware tool access and tool-group boundaries.
• Data and security layer: PostgreSQL (+ replica/PgBouncer), Redis, Kafka, and KeyVault providers.

---

🚚 Deployment Options

Langboard ships as a containerized stack with core services for server, ui, api, socket, graph, and celeryworker.

• Data and messaging services: PostgreSQL, PgBouncer, Redis, Kafka.
• Optional services: OpenBao (KEY_PROVIDER_TYPE=openbao-local) and Ollama CPU/GPU profiles.

---

🆚 Differentiation

• vs raw AI libraries: Langboard adds a collaborative workspace and governance controls, not just SDK primitives.
• vs standalone graph builders: Langboard keeps external Langflow compatibility while adding internal LangGraph execution, user roles, API key governance, and MCP policy boundaries.
• vs generic automation tools: Langboard is built around board collaboration plus AI execution, not only background jobs.

---

📜 License

Langboard is distributed under a source-available license inspired by Elastic License v2.

• ✅ Internal use: free and unlimited.
• ❌ No SaaS resale: cannot be offered as a hosted service without a commercial license.
• ⚖️ Commercial license required: for SaaS, resale, or bundled enterprise offerings.
• 🌐 GitHub Sponsors support is voluntary and does not replace a commercial license.

See LICENSE for full details.

---

🤝 Contributing

Contributions are welcome!

• Fork the repository and submit pull requests.
• Report issues and request features.
• Participate in discussions and share workflow templates.

Please follow our contribution guidelines.

---

❤️ Sponsorship

Langboard development is sustained through GitHub Sponsors.

👉 Sponsor Langboard on GitHub

---

📧 Contact

For commercial license inquiries, enterprise deployment, or partnerships, please reach out to:

📩 yamon@yamon.io