The wiki
This is what day thirty looks like.
A month of voice memos, screenshots, design critiques, and conversation notes. Stash crystallizes them into a structured wiki — written from your own captures, kept current, readable by every AI you use.
"LLM knowledge bases as an example of something that was impossible with classical code because it's computation over unstructured data (knowledge) from arbitrary sources and in arbitrary formats…"
Stash's wiki. About itself.
Not a mockup. This is what Stash actually thinks about Stash — synthesized from 110 captures over five months. No tagging, no filing. The page below updates itself every time something new is captured.
Stash is the personal context layer for AI. It provides invisible infrastructure for ambient capture and graph-shaped knowledge, emphasizing user ownership and data alignment over venture-backed growth models. Built to feed context to Claude, ChatGPT, and beyond — ensuring portability across different tools1.
Core philosophy & positioning
Stash views itself as a "new horizon" product — functionality that was impossible before LLMs1. It serves as the unification point for scattered context across digital silos, allowing any AI to read from one substrate2. Identity: durable product, durable substrate — invisible infrastructure built for longevity, not a venture-backed growth model3.
Knowledge graph & enrichment
Core architecture: graph-shaped knowledge infrastructure described as a "wiki that writes itself"4. The enrichment pipeline processes captures to extract entities, topics, and relationships — used to build and update wiki pages without tagging on the user's part5. A "dreaming pass" runs overnight to generate a daily brief, mirroring recent activity and offering wiki links for deeper exploration6.
Cross-tool by design
Stash exposes itself over the Model Context Protocol — every AI client that speaks MCP reads from and writes back to the same substrate7. Claude, ChatGPT, Codex, Cursor, and Microsoft Copilot are the current canonical clients; the protocol is open. The wiki you're reading was synthesized in part from threads across multiple AI tools, captured back to Stash via MCP.
References
- 1. Karpathy at Sequoia AI Ascent 2026 named LLM knowledge bases as a "new horizon"...
- 2. Stash as Gateway — the hub-and-spoke architecture thesis, crystallized...
- 3. Stash is not a startup. It's a durable product. A durable substrate. Invisible infrastructure...
- 4. Karpathy was right. Wikis are a really good feature for any knowledge base...
- 5. Stash needs to have a nightly job. It needs to synthesize all of the food for thought...
- 6. A "dreaming pass" runs overnight to generate a daily brief, mirroring recent activity...
- 7. Stash as gateway — hub-and-spoke architecture, MCP as the integration surface for cross-tool reach...
Tyler wrote none of this. Stash did. From 110 captures and counting.
How it got there
Capture, understand, synthesize, surface.
Four motions that turn raw captures into the wiki you just read. None of them require you to organize, tag, or file anything.
01 · Capture
Drop it in.
Voice on the walk. Screenshot of the chart. Email forwarded to capture@stash.bar. AI conversation pushed via MCP. Five ways in. One place.
02 · Understand
Stash reads it.
Every capture runs through the enrichment pipeline. People, projects, companies, and ideas get pulled out. New entities get their own page; existing ones get linked. No tagging on your part.
03 · Synthesize
Wikis write themselves.
Pages emerge over days, written from your own captures. Sections, infoboxes, references. Every claim traces back to the moment it was said.
04 · Surface
Every AI reads it.
Stash exposes an MCP server. Claude, ChatGPT, Codex, Cursor, Copilot — anything that speaks MCP — reads from your Stash and writes back. Every session adds to it.
And the rest of your life
Not only for software projects.
Kitchen renovations, vacations, kids' schedules, your parents' health. Anything you keep partial track of in your head — Stash holds it whole.
A kitchen renovation being planned for the house on Elm Street. Initial quotes came in higher than expected, leading to a phased approach starting with cabinets and countertops. Mom suggested checking Habitat ReStore for hardware — saved 40% on her remodel that way. Three contractor quotes total; Mike Chen at $41k was the only one to suggest phasing.
Captured: voice memos, a forwarded quote from Ferguson, a missed call from Mom, three contractor meetings. Stash made the page.
Your wiki starts building from your first capture.
Whatever you've got piling up — half-finished projects, scattered notes, the thing your collaborator told you last Tuesday — Stash takes it. The wiki writes itself. Every AI you use catches up.
Join the waitlist
No spam.