Most teams that want everyone using the same AI agent reach for a cloud drive — Dropbox, Google Drive, OneDrive — drop the agent files in a shared folder, and sync. It works, but it ships your prompts, your configurations, and often your clients' data through a third party you do not control. ITECS took a different path. We run our own file-sync server and use it to share the same custom AI agents with every employee, without a single client file leaving our security boundary.
ITECS shares App-enabled AI agents — the kind that run natively inside Codex and Claude Cowork — by syncing them through a self-hosted Seafile server in our own datacenter. Every employee on Windows or macOS gets the same agents in their project folders automatically, and no client data ever touches a public cloud file service like Dropbox.
The Problem With Sharing AI Agents Through a Cloud Drive
An AI agent is only useful to a company if everyone can use the same one. When one technician builds a strong documentation agent, the whole team should inherit it the next morning — not rebuild it from scratch. The obvious way to distribute it is a shared cloud folder. That convenience carries a quiet cost.
Consider a 60-person managed services firm that supports dozens of clients. An engineer builds an AI agent that reads a client's network documentation and drafts the missing runbooks. To share it, the team drops the agent — and the client documentation it reads — into a Dropbox folder. Now a third-party vendor stores network diagrams, admin procedures, and onboarding notes for every client. One misconfigured share link or vendor breach exposes data the firm contractually promised to protect.
That is the trade most teams never examine. The agent files are not the sensitive part — the client data they touch is. Sync that through a public cloud and you have moved your clients' information outside your control to save a few minutes of setup.
What ITECS Built: One Shared Library of AI Agents
ITECS solved this with infrastructure we own end to end. The core is Seafile, an open-source file-sync platform, running on a Linux server inside our Dallas datacenter. It does what Dropbox does — keep folders identical across many computers — except we host it, we hold the encryption keys, and the data stays on our hardware.
Three pieces work together. Seafile handles secure file sync and version history across the workforce. Codex and Claude Cowork are the AI tools our employees run locally; both support App-enabled agents, meaning an agent is defined by files — instructions, skills, and scripts — that live in a project folder. The Seafile client on each Windows and macOS machine keeps those project folders identical for everyone. If your team is new to these tools, start with our guide to Claude Cowork for small business.
When an engineer improves an agent on their laptop, Seafile syncs the change to our server, then out to every other endpoint. The next time a colleague opens Codex or Cowork in that project folder, they run the updated agent — no install, no copying, no manual versioning. The whole company shares one evolving library of AI agents the same way a team shares a set of documents. We covered the broader pattern in building a shared repository of agentic skills.
Meet DOCBOT: The Agent That Manages Client Documentation
The clearest example is an agent we call DOCBOT. It manages client documentation — the runbooks, standard operating procedures, onboarding and offboarding checklists, and knowledge-base articles a managed services team lives on. DOCBOT lives in a synced project folder, so every technician runs the same version.
A technician asks DOCBOT to do three kinds of work. It adds new documentation, turning a finished project or a support ticket into a clean SOP. It updates existing records when a client's environment changes, so the docs match reality. And it discovers gaps, flagging the runbooks, onboarding steps, and knowledge-base articles a client should have but does not. Documentation stops being the task everyone skips and becomes a prompt away.
In practice, a technician finishing a firewall replacement tells DOCBOT what changed. DOCBOT drafts the updated network runbook, adds the offboarding step for the old device, and checks whether the client's onboarding guide still references hardware that no longer exists. A human reviews and approves. The documentation that used to lag weeks behind the work now keeps pace with it. DOCBOT is one pattern of the AI knowledge base work we deliver for clients, applied to our own operations.
How the Shared-Agent System Works
Setting up shared, self-hosted AI agents follows four steps. ITECS runs the same playbook for clients who want their own private version.
Step 1: Stand up a self-hosted sync server. We install Seafile on a hardened Linux server inside the datacenter, behind the firewall, with encrypted storage and scheduled backups. Nothing depends on a third-party cloud.
Step 2: Define agents as files in project folders. Each AI agent — its instructions, skills, and scripts — lives in a versioned project folder. Because Codex and Claude Cowork read agents from the folder, the folder is the agent.
Step 3: Sync to every endpoint. The Seafile client on each Windows and macOS machine keeps those folders identical. New hires get the full agent library on day one by signing in.
Step 4: Govern changes like code. Edits sync with version history, so we can see who changed an agent and roll back a bad update. Sensitive actions still require human review before they run.
Why Seafile, and Why Self-Hosted
We chose Seafile deliberately. It is open-source, so we can audit exactly how it handles our data instead of trusting a closed vendor black box. It is self-hosted, so the server, the storage, and the encryption keys are ours. And it keeps client data inside our boundary, which is the whole point for a firm that signs confidentiality agreements with every client.
Public cloud sync tools like Dropbox, Google Drive, and OneDrive are excellent products, but they place a third party between us and our clients' data. For a managed services provider, that is a contractual and security liability we choose not to carry. Self-hosting trades a little convenience for full custody of sensitive information.
Security and Data Boundaries
Keeping data in-house is not automatically secure — it has to be operated that way. We treat the agent platform like any production system: scoped access per employee, encrypted storage, audited changes, network segmentation, and human approval gates on anything that touches client data. We align the controls to the NIST AI Risk Management Framework, the U.S. standard most enterprise auditors now use for AI accountability. Before any client's data feeds an agent, we run a data and AI readiness audit to confirm it is classified, access-controlled, and safe to use.
This is the same discipline behind ITECS' custom AI agents for clients — private environments, no third-party model training on your data, and the backing of a cybersecurity practice ITECS has run since 2002. For teams adopting these tools, structured AI training makes sure people actually use the approval gates instead of working around them.
What It Costs and the ROI of Shared Agents
The economics are straightforward. A self-hosted sync server runs on hardware most firms already own, plus the open-source Seafile software, which carries no per-seat license. The real return is operational: one engineer's best agent becomes every engineer's tool overnight, and documentation that used to be skipped gets done.
ITECS builds these systems on its standard model — hourly consulting or prepaid retainer hours with tracked usage, no minimum monthly commitment and no expiration, plus a flat fee for scoped builds like a private agent platform or a custom workflow automation. The ROI shows up as hours returned to billable work and as risk removed, because client data never leaves your control. When you want the same setup for your team, talk to the ITECS team.
Want a private, self-hosted AI agent platform that keeps client data inside your walls? Learn about our Custom AI Agents service or schedule a free AI assessment.
About The Author
The ITECS Team
ITECS helps Dallas business leaders adopt practical AI with the security, documentation, training, and operational discipline expected from an established managed technology partner.
Sources And Trust Signals
This article is based on ITECS implementation experience and the public resources below.
The open-source file-sync and version-control platform ITECS self-hosts on Linux to keep AI agent project folders identical across every employee endpoint.
The U.S. standard ITECS aligns its self-hosted AI agent controls to — scoped access, audit logging, and human approval gates that enterprise auditors expect.
ITECS service page for private, governed custom AI agents with scoped credentials, audit logging, and no third-party model training on your data.
The documentation and knowledge-base service the DOCBOT pattern is built on — runbooks, SOPs, onboarding and offboarding content kept current.
The data classification and access-control review ITECS runs before any client data feeds a self-hosted AI agent.
ITECS guide to Claude Cowork — one of the App-enabled AI tools employees run locally on top of the shared, synced agent library.