AgenticOS: Open-Source AI Teammates for Your Enterprise

Overview

Vstorm has released AgenticOS, an open-source agent layer designed to help enterprises deploy and manage shared AI teammates across their infrastructure. Licensed under Apache-2.0, the platform allows organizations to run AI agents on servers they control, addressing growing corporate demand for data sovereignty and security in artificial intelligence adoption.

The release positions AgenticOS as a solution for companies looking to move beyond isolated chatbots toward integrated, collaborative AI systems. By building on the Pydantic AI framework, the platform offers a structured environment where agents can be built, taught company-specific knowledge, and connected to internal tools without relying on third-party cloud processing for sensitive data.

"Growing a company used to mean hiring more experienced people. Here their procedures become skills, their standing facts become context, and every agent works from both." — Vstorm

Core Philosophy: Institutional Knowledge as Code

The primary value proposition of AgenticOS is its approach to knowledge management. Rather than training large language models on proprietary data, which can be costly and risky, the platform allows organizations to inject their existing expertise into agents through plain-language documentation.

When a skilled employee leaves or a procedure changes, updating a simple Markdown file ensures that every agent referencing that information is immediately updated. This creates a dynamic system where institutional knowledge is version-controlled and auditable.

Key aspects of this approach include:

• Skills: Defined procedures and workflows that agents follow, such as invoice verification or procurement rules.
• Context: Static factual data, such as preferred supplier lists or approval thresholds, that informs agent decision-making.
• Memory: Short-term and long-term recall capabilities, allowing agents to remember past interactions and outcomes.

This structure means that an agent’s behavior is governed by human-readable files, making it easier for non-technical stakeholders to understand and modify how AI teammates operate.

Capabilities and Integrations

AgenticOS comes with 26 built-in capabilities that can be toggled on a per-agent basis. These are organized into functional categories ranging from web interaction to safety guardrails.

Functional Toolsets

Agents can be equipped with tools for:

1. Web Interaction: Performing searches, reading web pages, and driving a real browser interface.
2. File and Code Processing: Running calculations, generating charts, creating images, and executing code in a sandboxed environment.
3. Reasoning and Collaboration: Planning complex tasks, thinking through problems step-by-step, and delegating sub-tasks to other specialized agents.
4. Safety and Limits: Redacting personally identifiable information (PII), managing long conversation lengths, and enforcing tool output limits.

Each capability acts as a card in the agent’s toolbox. Engineers can also add custom capabilities using typed Python, extending the system far beyond its default offerings.

Model Agnosticism

A significant feature of AgenticOS is its independence from specific AI providers. The platform supports 26 model providers, including major commercial options like Anthropic, OpenAI, Google Gemini, Mistral, DeepSeek, Azure OpenAI, and AWS Bedrock. It also supports local deployments via Ollama, catering to enterprises with strict offline requirements.

This flexibility allows organizations to choose the best model for each task—using expensive, high-reasoning models for complex strategic analysis and cheaper, faster models for routine queries—while maintaining a unified management interface.

Enterprise Connectivity

The platform integrates with over 4,700 MCP (Model Context Protocol) servers, enabling agents to connect to a vast ecosystem of external applications. Native support is provided for common enterprise tools such as HubSpot, Notion, Slack, Mattermost, and Telegram.

Agents can be reached through:

• Web chat interfaces
• Slack, Mattermost, or Telegram channels (where mentions trigger agent responses as the sender)
• HTTP APIs for custom integrations
• Website widgets

Deployment and Workflow

AgenticOS is designed to be self-hosted, ensuring that all data remains within the company’s infrastructure. While the software is free to run under the Apache-2.0 license, Vstorm offers rolled-out deployment assistance for organizations that require professional support.

The typical workflow for deploying an agent follows a seven-step process:

1. Build: Naming the agent, writing instructions in plain language, and selecting required tools.
2. Teach: Uploading handbooks and documents, which are parsed, indexed, and cited during responses.
3. Connect: Signing into external apps and configuring which tools the agent may access.
4. Chat: Testing the agent in a web interface, dropping in files, and reviewing its step-by-step reasoning.
5. Slack: Integrating the agent into communication platforms for daily use.
6. Automate: Scheduling routine tasks based on triggers like calendar events, GitHub issues, or webhooks.
7. Share: Publishing dashboards and agent versions with stable links for team access.

Subagent Coordination

One of the more advanced features is the ability for a "lead" agent to delegate parts of a task to specialized subagents. For example, a Chief of Staff agent might ask a Finance Analyst agent to review a budget discrepancy and a Sales Researcher agent to update contact information. The lead agent then synthesizes these inputs into a final brief.

This mimics human team dynamics, where different specialists contribute to a larger project. Users can observe the interactions between agents, seeing exactly who was asked and what each one contributed, ensuring transparency in automated workflows.

Target Audience and Availability

The platform is aimed at enterprises of all sizes that need to deploy AI while maintaining control over their data and intellectual property. It is particularly relevant for organizations in regulated industries, such as finance and logistics, where audit trails and data residency are critical.

Vstorm provides 29 tutorials to help teams get started, covering sample inputs, configurations, and validation checks for common use cases. These resources allow companies to begin with a single repetitive task and expand from there, gradually building a roster of AI teammates tailored to their specific operational needs.