0ad6ee3156
* Add blog post: Adding Long-Term Memory to LangGraph and LangChain Agents * blog: update langgraph post date to 2026-03-24 and add cover image * blog: fix claude-code-telegram filename to match frontmatter date (2026-03-25) * blog: set claude-code-telegram date to 2026-03-23 * blog: fix date timezone offset by adding T12:00 to all post dates * ci: trigger fresh CI run * blog: fix broken docs link (routeBasePath is /)
51 lines
2.7 KiB
Markdown
51 lines
2.7 KiB
Markdown
---
|
|
title: "What's New in Hindsight Cloud: Document File Upload"
|
|
authors: [benfrank241]
|
|
date: 2026-03-09T12:00
|
|
tags: [hindsight-cloud, release, memory]
|
|
hide_table_of_contents: true
|
|
---
|
|
|
|
[Hindsight Cloud](https://ui.hindsight.vectorize.io/signup) now lets you upload files directly to any memory bank. PDFs, Word documents, PowerPoint presentations, Excel spreadsheets, images, and plain text files are all supported.
|
|
|
|
<!-- truncate -->
|
|
|
|
## Two Extraction Methods
|
|
|
|
When you upload a file, you choose how Hindsight processes it:
|
|
|
|
**Standard extraction** uses [Markitdown](https://github.com/microsoft/markitdown) to pull text from your document. It is free and works well for text-heavy files like PDFs, Word docs, and plain text.
|
|
|
|
**Enhanced extraction (Iris)** uses AI-powered processing to understand the structure and meaning of your document. This is useful for complex layouts, scanned documents, and images. Enhanced extraction incurs per-token charges.
|
|
|
|
Both methods extract the content and store it as structured memories in your bank, ready for recall and reflect queries.
|
|
|
|
## How It Works
|
|
|
|
1. Open any memory bank in [Hindsight Cloud](https://ui.hindsight.vectorize.io/signup)
|
|
2. Click the upload button and select your files
|
|
3. Choose Standard or Enhanced (Iris) extraction
|
|
4. The Document Operations panel tracks progress with status indicators
|
|
|
|
Once processing completes, the extracted content becomes part of your memory bank. Your agents can recall and reflect on the document contents just like any other memory.
|
|
|
|
## Supported File Types
|
|
|
|
- **PDFs** -- reports, whitepapers, research papers
|
|
- **Word documents** (.docx) -- meeting notes, specs, proposals
|
|
- **PowerPoint presentations** (.pptx) -- slide decks, training materials
|
|
- **Excel spreadsheets** (.xlsx) -- data tables, financial reports
|
|
- **Images** (.png, .jpg) -- screenshots, diagrams (Enhanced extraction recommended)
|
|
- **Plain text** (.txt, .md) -- logs, notes, documentation
|
|
|
|
## What Else Is New
|
|
|
|
This update also follows two other recent additions:
|
|
|
|
**Bank-scoped API keys** (March 3) let you restrict API keys to specific memory banks. This is useful for multi-tenant setups where each customer or agent should only access its own memories. Unauthorized access attempts return a 403 error.
|
|
|
|
**MCP support** (February 13) added [Model Context Protocol integration](/blog/2026/03/04/mcp-agent-memory) for AI clients like Claude, Cursor, and VS Code. You can connect in single-bank mode for a dedicated agent memory or multi-bank mode for cross-bank operations, with 30 tools available out of the box.
|
|
|
|
## Get Started
|
|
|
|
File upload is available now in [Hindsight Cloud](https://ui.hindsight.vectorize.io/signup). Upload your first document and try a recall query against it to see the results.
|