ONLYOFFICE DocSpace 3.7: Generate Full Documents with AI in Seconds

If you have been watching the open-source office suite space, you know ONLYOFFICE has been steadily building something interesting. Their DocSpace platform, launched in 2023, sits in that sweet spot between Google Workspace and Microsoft 365, but with one crucial difference: you can actually self-host it. For teams and individuals who want format compatibility without handing their data to a tech giant, that matters. Version 3.7 just dropped, and it brings some genuinely useful upgrades, especially around AI integration and form handling.

I have been running DocSpace in a homelab container for about six months. I use it for client-facing documents where I need Microsoft Office compatibility and the freedom to keep the data on my own server. The 3.7 release is the first one in a while that actually changed my daily workflow.

The headline feature: AI-generated files

The biggest addition in 3.7 is the ability to generate a real DOCX, PDF, or PPTX file directly from an AI chat prompt. You ask the DocSpace AI agent to create a Word document, a PDF form, or a PowerPoint presentation, and it produces an actual file that opens immediately in the editor. It is not text pasted into a blank document. It is a properly formatted file with structure, headings, and layout.

For anyone who regularly churns out reports, proposals, or presentation decks, this cuts out the tedious setup phase. You still need to review and refine what the AI produces, but starting from a complete draft beats staring at a blank page. I generated a project status report on Monday and cut maybe twenty minutes off the usual “set up the template, write the headings, copy in last week’s numbers” routine. The generated file came out with the right section breaks, a working table of contents, and consistent heading styles, which is more than I expected from a first-pass generation.

The file generation works through the same AI agent chat interface that DocSpace introduced in version 3.6. The agent reads your prompt, calls the configured model, and produces the file in the right format. You can also configure custom AI providers if you are running local models through Ollama or a service not on the built-in list. I tried a few different prompt styles to see how the output quality varied, and the model choice made a bigger difference than I expected. Anthropic gave me the cleanest prose. DeepSeek was faster but produced rougher layouts. Local Ollama models handled simple forms well but struggled with longer presentations.

Seven AI providers, real flexibility

The new release added three more AI providers to the lineup. DeepSeek, xAI, and Google AI joined the existing roster, which already included Anthropic, OpenAI, OpenRouter, and Together AI. That is seven providers total, plus the custom option.

The practical benefit is not “look at all the logos.” It is that you can match the model to the task. I route long-form writing through Anthropic because the prose quality is what I want for client deliverables. I use a local Ollama instance for summarizing internal meeting notes, because the data is sensitive and the quality bar is lower. DeepSeek is my fallback when I want cheap inference for bulk document processing. The point is, I can pick.

API keys are required for each provider you connect, and DocSpace stores them in the application config, not in the documents themselves. The setup is a settings page, not a YAML file, which is a small thing I appreciate.

Form filling got the work it needed

The form filling improvements in 3.7 are less flashy than the AI features, but they are probably more immediately useful for teams that use DocSpace for intake workflows. Form filling rooms now support in-room PDF creation and editing, which means you can build a PDF form, distribute it to a room of collaborators, and collect responses without leaving the room.

The new “Start filling” mode makes collaborative form completion smoother. Before, multiple people editing the same form would step on each other’s changes. Now the room tracks who is editing which field, and responses sync to a connected XLSX spreadsheet as they come in. For HR teams, client intake workflows, or any team that runs a lot of structured data collection, this is the kind of quality-of-life upgrade that pays for itself in a week.

What I would skip, and what to plan for

There are a few things to know before you upgrade. The AI file generation feature requires you to connect at least one AI provider, which means an API key and an active subscription somewhere. If you are running DocSpace specifically to avoid that kind of vendor dependency, you can skip this feature, but you will not get the headline benefit of the 3.7 release.

The new form filling room features work best with a recent browser. Older versions of Edge and Safari have rendering bugs in the PDF editor. The team has acknowledged this and the fixes are landing in 3.7.1, but if your team is on locked-down corporate browsers, hold off for the point release.

For self-hosters, the Docker image sizes have crept up. The 3.7 image is about 40% larger than 3.6, mostly because of the bundled AI integration libraries. If you are running DocSpace on a low-resource VPS, plan for an extra gigabyte of disk and a few hundred megabytes of memory.

A few things that have held up under testing

• Set up a custom provider pointing at your own Ollama instance first. You can add cloud providers later, but starting local keeps your data on your own hardware by default.
• Use the room tracking for any form that goes to more than three people. The “who is editing which field” feature prevents the kind of overlapping edits that used to silently corrupt form responses.
• Back up the document store before the upgrade. The migration is clean, but the database schema change in 3.7 is large enough that a backup gives you a clean rollback path.
• Check the new image size budget. On a 2GB VPS you will need to bump to 3GB or 4GB to handle the new image plus the AI service overhead.

Trade-offs

DocSpace 3.7 is not a free lunch. The AI features require you to be comfortable sending prompts and document context to a third-party model provider. If you are using the built-in providers, your data leaves your server. The custom provider option helps, but only if you are willing to run your own model infrastructure.

The form filling improvements are real, but they assume a specific workflow: structured intake, multiple collaborators, spreadsheet export. If your team does not run that kind of process, the form features will not change anything for you.

The migration from 3.6 to 3.7 took me about twenty minutes on my homelab setup, including backing up the database and the document store. The upgrade was clean, and the rollback path was straightforward because I had the old image tag still in my Docker Compose file.

If you are already self-hosting DocSpace, this is a clean win. The AI features are genuinely useful, the form filling improvements are real, and the migration is low-risk. If you are evaluating DocSpace for the first time, the 3.7 release is a good reason to give it another look, especially if Microsoft Office compatibility and AI integration are both on your requirements list.

If you only do one thing from this article, set up a custom AI provider in DocSpace pointing at your own Ollama instance before you connect any of the built-in providers. You can always add the cloud providers later, but starting with local models keeps your data on your own hardware by default.