Tensorlake January Updates: Agentic OCR for Tables & Charts + New Pricing That Scales With You

Engineers building production document workflows - we've shipped two powerful new capabilities and restructured pricing to better serve you at every stage.

🧩Agentic Table Merging: Fragmented tables that span pages or across multiple columns? Document AI now reconstructs them into single tables by merging them.

📊Agentic Chart Extraction: Turn static chart images into structured JSON language models can use for analytics using code sandboxes.

💰New Pricing: 70% off for startups, predictable Pro pricing as usage scales, and lower rates for enterprise scale.

Agentic Table Merging

Ever asked an LLM "What was the total balance at the end of the year?" on a 200-page financial PDF? The processing pipelines to handle such queries require a lot of manual effort in stitching tables back using heuristics. The problem: most parsers treat multi-page tables as separate fragments.

Our new Agentic Table Merging fixes this. Instead of relying on geometry alone, we use a specialized agent that analyzes content and context to determine if table fragments belong together.

What it handles:

• Cross-page merges: Tables that continue across page breaks, even with repeated headers and footers

• Same-page merges: Tables split into columns that logically belong together

Why it matters:

• Better chunks for RAG pipelines

• More accurate retrieval on queries that need the whole table

• Stronger LLM reasoning — no more missing rows or confused totals

Turn it on with table_merging=True in the SDK or API.

👉 Read the full blog →

Agentic Chart Extraction

Charts in PDFs are hard to use in analytics pipelines. Our new Agentic Chart Extraction transforms static chart images into dynamic, structured JSON — ready for plotting, analytics, or LLM consumption.

Key capabilities:

• Chart type detection: Line, bar, scatter, pie — high accuracy across common types

• Data series extraction: Returns structured category/value pairs and coordinates

• Robustness: Handles multi-series charts, varying axis scales, and 50+ point dense scatter plots

• Directly plottable: JSON outputs conform to standardized schemas — feed them straight into your plotting library

Available now with all OCR models.

👉 Read the full blog →

We're restructuring pricing to do two things: support early-stage companies building with document AI, and offer more scalable rates for our largest customers as they grow.

What's Changing

What's Adjusting (February 2026)

• Base OCR pricing increases to $0.03/page (from current rates)

• Summarization & Structured Extraction will cost $4 per million tokens — for most use cases, this adds roughly ~1 cent per page

Why We're Making These Changes

Our largest customers are processing millions of pages. They need pricing that decreases with volume. Early-stage companies need room to build.

Our on-demand API requires zero commitment from users — no contracts, no minimums. But we still carry the storage and compute costs to keep it available. The base rate increase lets us continue offering true $0 upfront pricing while covering those costs.

These changes fund continued development of agentic document intelligence while making the economics work at every scale.

Questions? Book a time and let's talk, or send us an email - [email protected] 

💡 Quick Links

• Try Agentic Table Merging →

• Try Agentic Chart Extraction →

Thanks for being a part of the Tensorlake Community. We are here to help you build Agentic applications and Document processing pipelines that work for your company!

The Tensorlake Team