AiFiler's Latest Upgrades: Execution Policies, Real-Time Insights, and CSV Support

The Headline Changes

Three weeks ago, we pushed four changes that directly affect how you work with documents every day. Goal-based AI agents now execute with execution policies and auto-resolve logic. The Intelligence tab dropped cost estimates from the user-facing view. File uploads now accept CSV and HTML files. And retrieval got smarter with Contextual Retrieval, a hybrid search system that understands what you're actually looking for.

If you've been using AiFiler since last quarter, you'll notice the difference immediately. If you're new, these features are why you picked us.

Goal-Based Agents Now Execute With Real-Time Insights

The biggest change is how agents work. Before, you'd set a goal (like "monitor all client contracts for renewal dates") and the agent would run in the background. You'd check back later to see results. That's fine for overnight jobs, but it's not how people actually work.

Execution policies change that. Now when you create a goal-based agent, you define:

• When it runs (immediately, on schedule, or on-demand)
• What it does (extract data, flag issues, generate summaries)
• How it resolves (auto-close if certain conditions are met, or wait for human review)

The real-time insights piece matters more. As the agent processes your documents, you see progress in real time. You're not waiting for a batch job to finish. You're watching it work.

Here's what that looks like in practice:

1. Open Intelligence (Ctrl+Shift+A on Windows, Cmd+Shift+A on Mac)
2. Click Create Goal and describe what you want monitored
3. Set your execution policy (e.g., "Run daily, auto-resolve if confidence > 95%")
4. Watch the agent process documents as they arrive

The agent learns from your feedback. If you override an auto-resolution, it adjusts. If you approve one, it gets more confident. This is the opposite of fire-and-forget automation.

Cleaner Intelligence Tab, Better Focus

We removed cost estimates from the Intelligence tab. Why? Because you don't need to see how much Claude costs per API call. You need to see what the agent found.

The tab now shows:

• Agent status (running, idle, paused)
• Last execution timestamp
• Results summary (5 contracts flagged, 12 documents processed)
• Next scheduled run (if applicable)

No noise. No pricing details. Just signal.

CSV and HTML Uploads—Finally

File support expanded. You can now upload:

• CSV files (parsed as structured data)
• HTML files (converted to readable text)
• Everything else you could before (PDF, DOCX, TXT, images, etc.)

This matters because a lot of data lives in CSVs. Spreadsheets. Exports from other tools. Now you can dump them into AiFiler and ask questions like "Show me all customers from Q4 with revenue > $50k" and the agent will parse the CSV, understand the structure, and answer.

HTML is useful for archived emails, exported web pages, and documentation that was saved as HTML instead of PDF.

To try it:

1. Click Upload in the sidebar
2. Select a CSV or HTML file (or drag it in)
3. AiFiler parses it automatically
4. Search or ask questions about it immediately

Contextual Retrieval: Smarter Search Under the Hood

This one is more technical, but it affects every search you do.

Before, AiFiler used keyword matching + vector search. That works for simple queries. But if you asked "Show me all contracts where we negotiated better payment terms," the system would struggle because "negotiated" and "payment terms" might not appear together in the text.

Contextual Retrieval changes that. It's a hybrid system that:

1. Understands intent from your query (you're looking for contract modifications, not just mentions of "terms")
2. Searches semantically (finds documents about payment negotiations, not just exact phrase matches)
3. Ranks by relevance (puts the most useful documents first, not the ones with the most keyword hits)

You don't need to do anything different. Just search like you normally do. The retrieval got smarter behind the scenes.

The technical detail: AiFiler now uses the Anthropic Files API for document storage and retrieval, which means documents stay in Claude's context window longer and retrieval is faster. If you're curious about the architecture, we documented it in our CLAUDE.md file in the repo.

What's Coming Next

We're working on two big things:

Deep Document Phase 1 will let agents understand relationships between documents. Right now, an agent can read a contract. Soon, it'll understand how that contract relates to your SOW, your invoice, and your communications with that client. Everything connected.

Knowledge graphs are next. Instead of just searching documents, you'll be able to visualize how information connects. Who works with whom. Which projects share vendors. Which contracts have overlapping terms. The graph builds itself as you upload documents.

Both ship in the next 6 weeks.

How to Get Started With These Changes

If you haven't tried the new agent features yet:

1. Open Intelligence and create a test goal (something simple like "flag all documents mentioning 'urgent'")
2. Set an execution policy (run immediately, auto-resolve if found)
3. Watch it work in real time
4. Upload a CSV file and ask a question about it
5. Notice how search results improved

These aren't small tweaks. They're the foundation for how AiFiler handles knowledge work going forward. Try them and let us know what breaks or what you'd build next.