OpenBot

Product Introduction

1. Definition: OpenBot is a fully customizable, self-hosted agent application platform. Technically, it is a containerized deployment (Docker/Helm) that provides isolated, role-based AI agents with dedicated environments (browsers, files, tools) to perform tasks within a company's secure infrastructure.
2. Core Value Proposition: It exists to solve the critical problem of securely and transparently integrating AI into sensitive business workflows. Its primary value is enabling businesses to deploy custom AI agents built on their own technology stack, using their choice of LLM models, on infrastructure they own and control, ensuring data never leaves their environment.

Main Features

1. Agent Channels & Isolation: Each AI agent operates in its own dedicated channel, analogous to a private thread with a colleague. This provides persistent context and history for each task. Crucially, each agent runs in its own container with its own files and signed-in browser session, creating a security boundary per agent to prevent credential or data leakage between roles.
2. Skills-Based Agent Roles: Agents are not generic chatbots but are configured for specific roles (e.g., Warehouse Analyst, Inbox Triage, Tool Operator). Admins grant each agent only the specific skills (like /research or tool access) and knowledge sources its job requires, following the principle of least privilege.
3. Secure Knowledge Retrieval: Agents can answer questions from connected company knowledge bases (Google Drive, OneDrive). The system enforces a "fail-closed" permission model: it only returns documents the user asking the question already has permission to access, and always cites sources. Vectors and documents are stored in the company's own Postgres database.
4. Computer Use & Tool Operation: For tools without a usable API, agents can sign in and operate the user interface directly. A key feature is the ability to watch the agent's screen in real-time beside the conversation and to "take the wheel" to intervene or correct course without breaking the task flow.
5. AG-UI Protocol Foundation: Built on the open Agent–User Interaction Protocol (AG-UI), which standardizes how agents interact with the user interface. This allows OpenBot to integrate with any agent framework that speaks AG-UI (LangGraph, Mastra, PydanticAI, custom code), avoiding vendor lock-in to a single AI orchestration stack.

Problems Solved

1. Pain Point: The security and compliance risks of using hosted AI assistants where sensitive company data leaves internal systems, and the inability to audit or control AI actions within business tools.
2. Target Audience: Chief Technology Officers (CTOs) and engineering leaders in mid-to-large enterprises requiring secure AI integration; DevOps/Security teams tasked with AI governance; product teams building custom AI-powered internal tools or customer-facing applications.
3. Use Cases: Secure internal helpdesks that answer from confidential policy documents; automated data retrieval and reporting from legacy systems via UI automation; AI-powered onboarding buddies for new employees; triaging and drafting responses for internal support inboxes; conducting competitive research with cited web sources.

Unique Advantages

1. Strengths & Limitations (Pros & Cons):

   • Pros:
     • Unmatched Data Control: Full self-hosting on company infrastructure with all data (vectors, chats, logs) residing in your Postgres. No shared tenancy.
     • Transparent & Auditable: Every agent action is pre-decided, recorded, and can be watched in real-time via screen sharing for UI tasks.
     • Security-First Design: Fail-closed retrieval, encrypted keys, per-tool allowlists for MCP, and agent-level isolation are core architectural principles.
     • Framework Agnostic: AG-UI protocol support means flexibility to use best-in-class or custom agent frameworks.
   • Cons:
     • High Implementation Overhead: Requires in-house DevOps expertise for deployment, maintenance, and scaling via Docker/Helm.
     • No Managed Service: The burden of security, updates, and model performance monitoring falls on the customer's team.
     • Initial Configuration Complexity: Setting up secure model endpoints, knowledge connectors, and role-based agent skills requires significant upfront technical planning.

2. Key Alternatives & Differentiation:

   • vs. Hosted Chat Assistants (e.g., ChatGPT Team, Microsoft Copilot for M365): OpenBot is self-hosted, so data never goes to a third-party cloud. It offers deeper, auditable integration with internal tools and UIs, whereas hosted assistants are limited to approved APIs and plugins.
   • vs. Open-Source Agent Frameworks (e.g., LangChain/LangGraph alone): OpenBot provides a complete, secure, production-ready application layer (UI, channels, permissioning, tool gateways) that you must otherwise build and harden yourself on top of a framework.
   • vs. Custom In-House Build: Differentiates by providing a pre-built, security-reviewed foundation with features like AG-UI, visual computer use, and channel management, significantly reducing development time and risk compared to a ground-up build.

Frequently Asked Questions (FAQ)

1. What is OpenBot, and what does it cost? OpenBot is a self-hosted AI agent platform you deploy on your own infrastructure. Pricing is not based on user seats but likely involves a commercial license from CopilotKit for enterprise use, support, and customization services, on top of your own infrastructure and model costs.
2. Can an OpenBot agent see documents I cannot? No. Its knowledge retrieval is "fail-closed." It strictly respects existing file permissions from sources like Google Drive. If you cannot access a document, the agent cannot retrieve or use it to answer your question.
3. Which systems can OpenBot connect to today? It natively connects to Google Drive and OneDrive for knowledge. For tools, it uses the Model Context Protocol (MCP) for resources and can perform "computer use" (UI automation) for systems without APIs. The specific MCP servers and UI automation targets are configurable by your team.
4. Do we need CopilotKit Intelligence to use OpenBot? No. OpenBot is architected to work with your choice of LLM providers (OpenAI, Anthropic, self-hosted models). CopilotKit Intelligence is a separate, optional managed service for certain AI features.
5. How does OpenBot compare to a hosted assistant like Grok Bot? The core difference is ownership and integration depth. Hosted assistants are shared, multi-tenant services with limited, API-driven tool access. OpenBot is a single-tenant, self-hosted platform that can perform deep, UI-level integration with internal tools and offers full data sovereignty and audit trails.