HeapThat has three layers. You capture anywhere. We normalise into clean Markdown and structured fields. You retrieve through a calm feed, a tag-graph, or your AI agents via MCP.
Six ways in: web composer, phone PWA, Telegram bot, browser voice recorder, email forwarding, and MCP. Anything your agents can call, your tools can email, or your phone can share to is a source. We don't build native integrations with every SaaS. Your other tools push to one of our channels.
Each item runs through a type-aware processor. Articles get cleaned to Markdown with Firecrawl. YouTube videos get their transcript via the Zyte proxy. Audio gets transcribed with Gemini. PDFs get text-extracted or vision-OCR'd. Then every item gets summarised with your editable prompt and auto-tagged.
A single-column feed, newest first. An archive page with full-text search and natural-language filters. A force-directed graph that maps your tags. A detail page per item with the summary, the cleaned Markdown, and any transcript or structured payload.
HeapThat exposes an MCP server. Claude Desktop, Claude Code, Cursor, and any MCP-aware agent can search the stash, retrieve any item, list your tags, and find upcoming contract renewals. Write-capable MCP at M11 lets agents add items, append notes, and tag on your behalf.
Contracts watch their own renewal dates. Wishlists watch their own prices. Subscriptions notify you before you forget you have them. Every agentic loop follows the same pattern: sense, extract, policy, actuator, learn.
Closed beta while we settle the core loop. Drop your email and we'll open a seat as soon as one frees up.
We use a small amount of analytics to understand which pages help and which don't. No ads, no profile selling. Pick what you're happy with.