AI runs through the whole platform: it splits and classifies documents, extracts fields, writes search queries and compares versions in the background. You can also use it directly on any document with the Docuflow AI functions assistant.
The AI assistant#
- Select documents on the Documents page, right-click and choose Docuflow AI functions.
- Choose what AI should do from the list (for example Summarize selected document(s)) or edit the question in Question(s) for AI. You can ask anything about the selected documents.
- Click Process. The answer appears in AI response, with word and character counts.
- Copy or Download the answer. Downloads are produced as PDF reports that include the documents and the question.
Comparing documents#
Select exactly two documents and choose Compare selected, or open Show versions and compare two versions of the same
document. AI reports what was added, removed and changed. With UseAutoCompare on, this happens automatically for every new version.
Where AI is used#
| AI function | Purpose | Triggered by |
|---|---|---|
| SplitDocument | Separates bundled PDFs into business documents, detects their type and extracts their content. | Background task |
| MapJson | Fits extracted content onto the document type's schema and indexes it. | Background task |
| Simplify | Cleans up a merged JSON template. | Template editor |
| Search | Turns a natural-language question into an OpenSearch query. | Search page |
| Document | Answers questions about and summarizes selected documents. | AI assistant |
| Compare / CompareVersionDiff | Compares two documents or two versions. | AI assistant, auto compare |
| Grammar / RewriteText / TextEditor | Proofreading and writing help (rewrite, change tone or style, expand, explain, shorten, summarize, translate) in rich-text editors. | Editor toolbar |
| LogSummary | Turns technical activity logs into readable summaries. | Activity logs |
AI Jobs#
AI Jobs (left menu) lists every AI request of your organization: status, time, user, function, processing priority and the AI response. Use the eye icon to see the full request and response. Background work runs at low priority; interactive requests run at high priority and are answered first.
Failed requests are retried automatically. Platform administrators can inspect failures under Admin → Platform logs → AI job failures and the raw calls under AI service logs. The RQ Dashboard (Platform management) shows the AI service's queues and workers.
Configuring AI (platform administrators)#
Under Admin → System administration → AI Settings:
| Page | What you configure |
|---|---|
| AI Providers | Available providers and models with their API keys: Claude, OpenAI, Gemini, DeepSeek, Ollama (self-hosted), vLLM (self-hosted) and Ollama Cloud. Use RQ queue is on for the self-hosted Ollama models: the Ollama server handles one request at a time, so their requests wait in the AI service's queue. Hosted providers and vLLM, which serves many requests at once, leave it off and process requests in parallel. vLLM names its models its own way (qwen3.8-27b, where Ollama has qwen3.8:27b). |
| AI Functions | The default provider and model for each AI function, so you can use a fast model for search and a stronger one for splitting. SplitDocument and MapJson run as separate background tasks, so each can use its own provider. |
| AI Prompts | The system prompts and the ready-made user prompts offered in the assistant, per function. |
The system parameter ShowSelectAIProvider adds a model picker to the assistant, which is handy for comparing models.
Cost#
AI operations are billed in credits based on processing time, with a higher rate for high-priority requests. See Credits & billing.