LLMWhisperer

@LLMWhisperer extract text from this 47-page contract PDF and format it so Claude can analyze the liability clauses

I've submitted your PDF for async extraction (whisper_id: wh_abc123). LLMWhisperer is processing 47 pages with layout-preserving mode enabled to maintain clause structure. Estimated completion: 90 seconds. I'll poll the status and retrieve the optimized text once ready, then analyze the liability sections you mentioned. This example assumes the conversion completes successfully; actual processing time varies by document complexity and current queue depth.

This showcases LLMWhisperer's core capability: converting complex PDFs into LLM-friendly text while preserving semantic structure. The async workflow requires polling with Check Whisper Status until completion. API key must have document conversion permissions. Large documents consume more page quota—check usage limits before batch processing.

@LLMWhisperer how many pages have I processed this month, and do I have enough quota left for 200 more scanned invoices?

Your LLMWhisperer account shows 3,847 pages processed in the current billing cycle (started March 1). Your plan includes 10,000 pages/month, leaving 6,153 pages available. Processing 200 scanned invoices at an average of 2 pages each would consume ~400 pages, well within your remaining quota. This example reflects typical usage data returned by the Get Usage Information tool; actual metrics depend on your account tier and processing history.

Demonstrates usage monitoring—critical for teams processing documents at scale. The Get Usage Information tool returns real-time quota data, helping you avoid mid-workflow failures due to limit exhaustion. No special permissions needed beyond API key access. Useful for capacity planning before bulk operations.

@LLMWhisperer set up a webhook at https://our-app.com/llm-results so we get notified when document conversions finish instead of polling

I've registered your webhook endpoint with LLMWhisperer. During registration, they sent a test payload to https://our-app.com/llm-results to verify connectivity—your server must respond with 200 OK to confirm. Once active, you'll receive POST notifications with whisper_id and status whenever async conversions complete, eliminating the need for status polling. This example assumes your endpoint is publicly accessible and configured to handle LLMWhisperer's payload format.

Shows webhook setup for event-driven workflows—more efficient than polling for teams processing many documents. Your endpoint must be reachable by LLMWhisperer's servers and return proper HTTP responses during registration. The auth token returned should be validated on incoming webhook calls. Use Delete Webhook to clean up unused registrations.

When LLMWhisperer makes sense for high-volume PDF intake

A 3-person legal ops team processes 40-60 vendor contracts per week, mostly scanned PDFs with mixed formatting. LLMWhisperer is the right call here because it handles the OCR and layout normalization before the LLM sees the text—no more garbled tables or missing clauses in the AI summary. The async mode with webhook registration means you drop the PDF in Slack, the extraction runs in the background, and the team gets a clean text block ready for clause extraction or risk flagging. The trade-off: if your contracts are already digital-native PDFs with clean text layers, you're paying for OCR you don't need. Check usage stats by tag to confirm you're not burning pages on documents a simpler tool could handle. For teams doing 200+ pages a week of scanned intake, this is the move.

Why LLMWhisperer works for one-off document questions

A 5-person support team gets ad-hoc questions about product specs buried in 80-page user manuals—PDFs that change every quarter. LLMWhisperer fits because you're not building a persistent knowledge base; you're converting a specific manual on-demand when a ticket references it, then feeding the text to the LLM for a targeted answer. The status-check tool lets you poll for completion without blocking the support agent, and the usage info tool keeps you honest about whether you're over-indexing on document conversion versus just maintaining a FAQ. The boundary: if the same 10 manuals get referenced daily, you should convert them once and store the text in a vector DB, not re-whisper them every time. For infrequent, high-page-count lookups where the source changes often, this is faster than manual copy-paste.

When async PDF processing beats real-time for research workflows

A 2-person nonprofit team reviews 15-20 foundation RFPs per month, each a 30-50 page PDF with eligibility criteria and reporting requirements. LLMWhersperer's async mode is the right fit because the team batches the research—drop all the RFPs into a folder on Monday, register a webhook to ping when extraction finishes, then spend Tuesday comparing requirements across the converted text. The tag-based usage stats let you track cost per grant cycle, which matters when you're on a tight budget. The limit: if you need instant answers during a live call with a program officer, the polling delay (even at a few seconds) breaks the flow. For batch research where you can wait 30 seconds per document and need clean text from scanned foundation PDFs, this is the cleanest path to LLM-ready input.