--- mw_bundle: 1 id: 9FATHAnw title: "Memory.Wiki Foundations" url: https://memory.wiki/b/9FATHAnw document_count: 9 updated: 2026-05-24T17:48:04.074Z analysis_generated_at: 2026-05-16T04:00:02.236Z analysis_stale: true source: "memory.wiki" --- # Memory.Wiki Foundations > What Memory.Wiki is, the three URL primitives, the memory architecture, the /memory.wiki skill across coding tools, the Bundle Spec, FAQ, and roadmap. Read in order, or paste the bundle URL into any AI for the whole pipeline at once. > ⚠ _Analysis may be stale — one or more member docs were edited after the last analysis run. Re-run the canvas to refresh._ ## Summary mdfy is a personal knowledge hub that uses public URLs as a universal interface for AI memory, solving the vendor lock-in problem of existing AI memory systems. The platform implements a three-tier architecture (Document/Bundle/Hub) where all content is stored as markdown accessible by any AI tool, with sophisticated retrieval capabilities including hybrid search and auto-synthesis. Unlike ChatGPT memory or Claude projects, mdfy knowledge is portable across tools, shareable with others, and owned by the user rather than trapped within a single vendor's ecosystem. ## Themes - Cross-tool AI portability - Public URL as universal interface - Human-AI readable knowledge - Vendor lock-in avoidance ## Cross-document insights - The URL is positioned as the 'contract' between mdfy and AI tools - a clever insight that leverages existing web infrastructure rather than requiring custom integrations - The skill-based integration approach using markdown instructions is more future-proof than binary plugins, as it works with any MCP-compatible tool - mdfy deliberately avoids becoming a chat UI or LLM provider, maintaining focus on publishing rather than inference - a strategic constraint that enables broader compatibility - The permanent URL commitment regardless of pricing tier is a key differentiator that builds trust and reduces switching costs ## Key takeaways - mdfy solves AI memory vendor lock-in by making knowledge portable through public URLs that any AI can read - The three-tier URL architecture (Document/Bundle/Hub) scales from single captures to entire knowledge bases while maintaining the same interface - Success depends on the URL-as-contract principle - if AI tools can fetch URLs, they can use mdfy without custom integration ## Open questions / gaps - No discussion of data export capabilities or what happens if users want to migrate away from mdfy - Missing details about enterprise or team collaboration features beyond basic sharing - Limited information about performance benchmarks or scaling characteristics of the retrieval system - No mention of versioning or conflict resolution for collaborative documents ## Notable connections - **doc:what-is-mdfy** ↔ **doc:mdfy-three-primitives** — The introduction's three primitives concept is fully detailed in the technical specification - **doc:mdfy-vs-vendor-memory** ↔ **doc:mdfy-memory** — The competitive positioning is backed up by technical details of how mdfy's memory system actually works - **doc:mdfy-skills-overview** ↔ **doc:mdfy-bundle-spec** — The skills integration relies on the bundle spec for consistent AI parsing across tools - **doc:how-mdfy-works** ↔ **doc:what-is-mdfy** — The architectural deep-dive explains the technical foundations for the user-facing features introduced in the overview - **doc:mdfy-faq** ↔ **doc:mdfy-vs-vendor-memory** — The FAQ addresses common comparisons that are detailed in the competitive analysis ## Concepts (this bundle) - **Public URL as Universal Interface** - **Cross-Tool Portability** - **Three-Scope URL Architecture** - **Human and AI Readable** - **Skill-Based Integration** - **Hybrid Retrieval System** - **Auto-Synthesis Pipeline** - **Permanent URL Commitment** ## Concept relations - **Public URL as Universal Interface** ↔ **Cross-Tool Portability** — enables portability - **Three-Scope URL Architecture** ↔ **Public URL as Universal Interface** — built on - **Skill-Based Integration** ↔ **Cross-Tool Portability** — supports principle - **Hybrid Retrieval System** ↔ **Auto-Synthesis Pipeline** — powers feature ## Documents ### 1. [What is Memory.Wiki](https://memory.wiki/what-is-mdfy) Memory.Wiki is a vendor-neutral URL-based personal knowledge system that lets users capture and organize AI conversations and documents in a single hub that any AI can read as context. The core innovation is replacing vendor-locked memory layers with a shareable markdown URL that works across ChatGPT, Claude, Cursor, and other AI tools without requiring SDKs or plugins. *sections:* The one-line version: A captured ChatGPT answer becomes memory.wiki/. Forty captures roll up into memory.wiki/hub/. You paste that hub URL into Claude, ChatGPT, Cursor, or C | Why a URL: Vendor memory layers (ChatGPT memory, Claude projects, Cursor docs) all live behind walls. They don't talk to each other, you can't share them, and you can't pa | Three primitives, one shape: Document at memory.wiki/ is one captured answer, paper, or transcript.; Bundle at memory.wiki/b/ is a curated grouping of docs around a topic, with its own URL.; Hub at memory.wiki/hub/ is your whole knowledge layer as a single deployable URL. | … ### 2. [Document, Bundle, Hub](https://memory.wiki/mdfy-three-primitives) Memory.Wiki uses a single URL primitive at three nested scopes (document, bundle, hub) to organize knowledge artifacts, each fetchable as markdown by any AI via HTTP. All three levels share the same viewer, fetch path, and embedding pipeline, allowing users to scale from individual documents to curated topic collections to their entire personal knowledge graph. *sections:* Document: memory.wiki/; AI fetch: same URL with Accept: text/markdown header (or /.md) returns the markdown body with frontmatter.; Embeddable: every public doc carries a 1536-dim vector and per-heading chunk vectors.; Editable: the owner edits in WYSIWYG; non-owners see the rendered viewer. | Bundle: memory.wiki/b/; Bundle Spec v1.0 conformant: stable shape, parseable by future tools.; Per-doc annotation: each member can carry a "why this is in the bundle" note.; Discoverable: owners can opt their bundle into the public /shared feed. | … ### 3. [How Memory.Wiki works](https://memory.wiki/how-mdfy-works) Memory.Wiki is a personal knowledge hub that captures AI conversation outputs, synthesizes them into structured wiki pages, and maintains a searchable schema so users can retrieve and reuse their knowledge across different AI tools through a simple URL interface that works universally across ChatGPT, Claude, and other AI platforms. *sections:* The problem: Every day, you get great answers out of an AI. Tomorrow, the AI doesn't remember any of them. The bookmark, the screenshot, the transcript copy you took, all of | Three layers: Memory.Wiki follows the same architecture Karpathy laid out (raw, wiki, schema), exposed as one product surface instead of three folders. | … ### 4. [Memory.Wiki Memory](https://memory.wiki/mdfy-memory) Memory.Wiki Memory is a public, human-readable memory layer that lets you capture answers from AI chats, organize them into a hub at a permanent URL, and share that hub with any AI via a simple HTTP recall endpoint that supports chunked retrieval, hybrid search (BM25 + vector), and privacy-enforced filtering. Unlike vendor-specific memory systems, it's not locked behind an SDK or MCP server and can be pasted into any AI as markdown context. *sections:* What "memory" means here: Every chat with ChatGPT, Claude, or Cursor produces useful answers. Tomorrow they're gone. The chat is closed, the share link rots, the next session has no idea | The surface (what you actually do): Paste a ChatGPT or Claude share URL into the editor.; /memory.wiki capture from inside Claude Code, Cursor, Codex CLI, or Aider.; Drop a PDF, DOCX, or transcript file.; Paste the hub URL into any AI. They fetch the markdown index and load your knowledge as context.; Hit the recall endpoint for question-targeted retrieval. Much fewer tokens, much higher precision: | … ### 5. [Memory.Wiki vs vendor memory](https://memory.wiki/mdfy-vs-vendor-memory) Memory.Wiki and vendor memory (like ChatGPT's built-in memory) serve different purposes: vendor memory personalizes within a single tool using automatic extraction, while Memory.Wiki enables human-curated knowledge to be shared across multiple AI tools and with other people via public URLs. The two are complementary, with vendor memory best for invisible preferences within one tool and Memory.Wiki best for substantive knowledge that needs reuse across tools and humans. *sections:* What vendor memory is good at: Stays inside the chat surface. ChatGPT memory is invoked transparently inside ChatGPT. You don't paste anything; the model just "remembers."; Auto-extracts atomic facts. "User prefers tabs over spaces." "User is building an iOS app." Small, structured, automatic.; Personalizes within one tool. The longer you use ChatGPT, the more it learns you in ChatGPT. | … ### 6. [/memory.wiki in your AI tool](https://memory.wiki/mdfy-skills-overview) Memory.Wiki is a skill installer that adds capture, bundle, and hub URL functions to AI coding tools like Claude Code, Cursor, Codex CLI, and Aider through simple markdown instructions. Once installed, users can save conversation segments, generate bundled documentation, and share their hub URL across any AI tool. *sections:* Install (one line per tool): bash | What it does, once installed: The same three verbs in every tool: | Why a skill rather than a plugin: A skill is plain markdown that the AI reads as instructions. No binary, no API key required (signed-out users get an anonymous capture cookie that they can clai ### 7. [Bundle Spec v1.0](https://memory.wiki/mdfy-bundle-spec) Bundle Spec v1.0 defines a stable markdown format for AI systems to fetch curated document collections from memory.wiki, with numbered sections, source citations, and privacy protections for restricted documents. The spec includes frontmatter metadata, per-document annotations for context, and is designed to enable accurate AI retrieval and citation across multiple platforms. *sections:* URL shapes: memory.wiki/b/<id> for the human-rendered bundle viewer; memory.wiki/b/<id>.md for the raw markdown payload; memory.wiki/raw/bundle/<id> (same as .md) | Frontmatter: yaml | Body shape: markdown | 1. <doc title>: Source: https://memory.wiki/<docId> | 2. <next doc title>: Source: https://memory.wiki/<docId> | Privacy: Member docs that became private after the bundle was created are filtered out at fetch time. The bundle structure stays intact (numbered sections), but unfetcha | … ### 8. [Memory.Wiki FAQ](https://memory.wiki/mdfy-faq) Memory.Wiki is a publishing platform designed so AI systems can read human knowledge bases as public markdown URLs, functioning as a memory layer across multiple AI tools rather than within a single vendor's ecosystem. It differs from personal knowledge bases like Notion by prioritizing AI readability, and from agent memory systems like mem0 by operating as a publishing layer rather than a backend API. *sections:* How is Memory.Wiki different from Notion / Obsidian?: Notion and Obsidian are great as personal knowledge bases for humans. Memory.Wiki is built so the AI can read it. The output is a public URL whose markdown is w | Do I need an account?: No. You can capture and share without signing up. Anonymous captures get a cookie that groups them, and you can sign in later to claim them all into your hub. | How does Memory.Wiki compare to mem0 / OpenMemory / Letta?: Those are backend memory layers for AI agents. They expose an API or MCP server, store atomic memories, and inject context automatically into one tool. Memory.W | … ### 9. [Memory.Wiki roadmap](https://memory.wiki/mdfy-roadmap-2026) Memory.Wiki is a document management and semantic search platform currently offering URL primitives at multiple scopes, multi-source capture, embeddings-based retrieval with vector and BM25 hybrid search, auto-synthesis features, and cross-AI verification at 100% accuracy. Upcoming features include auto-re-embedding on metadata changes and a cross-encoder reranker, while the team is deliberately deferring mobile apps and team accounts in favor of focusing on core functionality. *sections:* … _Digest view — follow any link above to fetch that doc's full markdown. Add `?full=1` to this URL for the concatenated payload._