The Change We Just Made
Three weeks ago, we flipped a switch. AiFiler now uses Claude as its primary AI provider instead of OpenAI. This wasn't a panic move or a cost-cutting measure—it was a deliberate choice based on months of testing and user feedback.
If you've been using AiFiler, you might not notice anything different when you open the app tomorrow. The interface is identical. Your documents are exactly where you left them. But the intelligence running behind the scenes? That's different now. And better.
Why We Switched
When we built AiFiler's AI system, we started with OpenAI because it was the obvious choice. GPT-4 was fast, reliable, and everyone knew how to use it. But as we built out features like the Universal Command (Ctrl+Shift+A on desktop), the Knowledge Graph analysis, and our document understanding system, we noticed something: Claude handled certain tasks with more precision.
Claude excels at:
- Long-form reasoning over documents. When you ask AiFiler to analyze a 50-page contract or synthesize insights across 20 related documents, Claude doesn't just skim—it actually reads.
- Structured output. Our intent handlers (we have 50+ of them) rely on consistent, predictable responses. Claude's output format is more reliable.
- Context awareness. Claude maintains better context across longer conversations, which matters when you're drilling into a complex document or building on previous search results.
We tested both systems side-by-side for three months. Claude won on accuracy. OpenAI won on speed. We chose accuracy because a fast wrong answer is worse than a slightly slower right answer.
What You'll Actually Notice
Here's the concrete difference:
Better document summaries. When you open a document and AiFiler generates a summary, Claude produces more nuanced breakdowns. It catches nuance that GPT-4 sometimes misses—the difference between "this contract is unfavorable" and "this contract is unfavorable because of these three specific clauses."
Smarter Universal Command routing. Universal Command is where you type natural language requests and AiFiler figures out what you want. "Show me all contracts from Q4 that mention liability" or "Create a summary of our vendor agreements." Claude's better at understanding intent, which means fewer misrouted requests and fewer "I didn't understand that" responses.
More useful ghost text hints. We recently shipped an auto-suggestions feature that shows you hints as you type in the editor. Claude generates these hints with better context—it understands what you're trying to write, not just what you've written so far.
How to Try It
You don't have to do anything. The switch is already live. If you have a preference and want to override it, here's how:
- Open Settings (bottom left, or Cmd+, on desktop)
- Navigate to AI Settings
- Look for the AI Provider dropdown
- You can manually select OpenAI if you prefer, but we recommend leaving it on Claude
That's it. The next time you use Universal Command, generate a summary, or ask AiFiler a question about your documents, you're using Claude.
What Else Changed Recently
We didn't just swap AI providers. We've shipped a bunch of other improvements:
Illustrated empty states. The Skills and Memory settings sections now have actual illustrations instead of blank gray boxes. Small thing, but it makes the app feel more intentional.
Knowledge Graph tooltips. When you have 5+ documents, AiFiler now shows you a tooltip suggesting you build a knowledge graph. This isn't pushy—it's just a helpful nudge when it actually makes sense.
Enhanced ghost text in the editor. The auto-suggestions setting now wires directly to the editor. Turn it on in AI Settings, and you'll see Claude's suggestions as you type—helpful context without being intrusive.
Better skeleton loaders. We replaced generic skeleton screens with content-shaped loaders in Memory and Skills settings. It's a small UX detail, but it makes the app feel more polished while things load.
The Architecture Behind the Scenes
If you're curious how we pulled this off without breaking anything: we built a unified AI client (lib/ai/client.ts) that abstracts away the provider. Instead of routes calling OpenAI directly, they go through this client, which handles routing to Claude or OpenAI depending on your settings.
We migrated three major routes first (as a test), then rolled out the change across the system. This meant we could flip the default without touching 50+ intent handlers or forcing users to reconfigure anything.
What's Coming Next
We're not done. We're currently working on:
- Multi-provider intelligence. Some tasks are better suited to Claude, others to OpenAI. We're building a system where AiFiler automatically picks the right provider for the right job.
- Streaming responses. Long document analyses currently wait for the full response. We're adding streaming so you see results as they arrive.
- Local model support. For teams that need to keep everything on-premise, we're exploring integration with local models like Llama.
The Claude switch is just the beginning. We're treating AI as a core system that should be flexible, not locked into one provider.
Try It Out
If you haven't used Universal Command recently, now's a good time. Open any document, hit Ctrl+Shift+A (or Cmd+Shift+A on Mac), and ask it something complex about your documents. Ask it to find patterns, summarize across multiple files, or generate a report.
The difference isn't dramatic—Claude isn't going to revolutionize how you work. But it's real. It's more accurate. It's more reliable. And that's what matters.
Enjoyed this article?
Get more articles like this delivered to your inbox. No spam, unsubscribe anytime.


