The reality of AI agents in self-hosted collaboration platforms
JumpCloud research reveals that 92% of organisations experienced AI-related incidents in the past year. Nextcloud offers a sovereign alternative by allowing users to host open-source AI assistants locally to prevent data sharing with third-party providers.
JumpCloud research indicates that 92% of organisations experienced at least one AI-related incident or exposure in the past 12 months. Organisations frequently give AI agents access to systems faster than they can secure them. More than half of organisations use AI agents that can change systems, permissions, or files, yet 59% do not follow the same governance rules for agents as they do for human users. This identity gap exists because only 41% of organisations using AI agents fully integrate them into the same identity and access management policies used for humans. Furthermore, 72% of organisations report a significant gap between what AI is allowed to access and what they can prove it actually accessed. More than one-third of organisations that investigated an AI-related concern were unable to determine what the AI had accessed. Research also shows differences in AI auditability across workplace software platforms, where just 11% of Google Workspace-primary organisations fall into the lowest auditability tier compared with 21% of Microsoft 365-primary organisations. Joel Rennich from JumpCloud states that every agent needs clear ownership, appropriate access, and controls that follow it throughout its lifecycle. Data-layer privacy platforms like Protecto and Skyflow focus on tokenization and masking as data moves to storage, but these tools do not determine whether an AI agent should be permitted to make a specific call at a specific time. Trussed AI addresses this runtime governance gap by enforcing policies on agent identity, least-privilege permissions, and tool approval workflows during execution. Because the agent places the user email address into the search query, the entire interaction becomes a disclosure to a third party that the company must list in its privacy policy to satisfy the transparency requirements of GDPR Article 13.
Nextcloud allows users to run an open-source AI assistant where they want it to be hosted. This prevents the sharing of data with third-party technology providers to train core AI models. The Nextcloud AI Assistant works across Nextcloud Hub to perform tasks in content writing, mail communication, and team chats. Users access the assistant via a single icon in the main navigation bar. Context Chat allows users to ask the assistant questions about files, collectives, and analytics data. The assistant can also perform agentic actions such as creating calendar events or sending a message in Nextcloud Talk. For text composition, Nextcloud Text allows users to summarize, reformulate, or change the tone of a document. Nextcloud also collaborates with European hosting companies like IONOS, OVHcloud, and PlusServer to provide AI-as-a-Service for users who do not want to host AI services themselves. Nextcloud is currently working with the German state of Schleswig-Holstein to build a digitally sovereign AI model for the public sector. This project uses anonymized training data to create standardized texts and optimize search for process-specific information. Nextcloud Mail summarizes email threads and generates reply suggestions based on the email text. Nextcloud also creates calendar events from emails and offers live transcriptions for meetings in Nextcloud Talk. The Context Write feature can derive context from existing files to convey their unique tone, wording, and structure to create new texts. Users can also generate documents, presentations, spreadsheets, and even PDFs using a simple prompt. Smart Picker allows users to access tools through the chat interface by typing a forward slash.
A functional AI usage policy must use clear data classification tiers. Organizations should separate public data from internal data and confidential data, such as client PII or proprietary source code. Using a consumer platform with confidential data is a risk because every prompt acts as a publication. You should implement human-in-the-loop controls for all HR functions, such as verifying AI-generated summaries before a manager uses them to make a decision. In Illinois, the HB 3773 statute prohibits AI from producing discriminatory effects across recruitment through termination. For companies in New York City, Local Law 144 requires an annual independent bias audit for automated employment decision tools. The Spanish data protection authority, the AEPD, published guidance stating that an agent will bring in recipients beyond the ones the processing already accounted for.
| Feature | Nextcloud | ownCloud |
|---|---|---|
| Encryption | Server-side and end-to-end | Server-side |
| 2FA | Standard feature | Add-on |
| Focus | Innovation and community | Stability and enterprise |
Self-hosting an assistant like LibreChat or Jan provides control over chat history and deployment. For those needing to chat with documents, AnythingLLM provides workspaces for organizing research. If a user needs a persistent agent for messaging channels, OpenClaw connects to Slack or Signal. Nextcloud provides more encryption options than ownCloud, including end-to-end encryption. Nextcloud also provides more frequent security updates and patches. Nextcloud is more intuitive and modern than ownCloud. To verify if an AI tool is truly private, turn off your wifi and see if the tool still works. If it still answers, nothing is leaving your machine. You might wonder if a local model like Command A+, which requires minimum hardware such as 4x B200 or 8x H100 GPUs, can handle the complex reasoning required for enterprise workflows.