{"slug": "show-hn-mindbase-an-llm-that-maintains-a-wiki-from-your-notes-and-papers", "title": "Show HN: MindBase – an LLM that maintains a wiki from your notes and papers", "summary": "MindBase, an AI research assistant that builds and maintains a persistent markdown wiki from user-supplied sources, is now in early access with beta feedback expected in Q3 2026. The tool implements Andrej Karpathy's LLM-Wiki pattern, using an LLM to read, cross-reference, and synthesize sources into structured wiki pages that compound knowledge across sessions, avoiding the re-derivation typical of RAG systems.", "body_md": "An AI research assistant that builds and maintains a wiki from your sources.Not RAG-in-a-vector-DB. A real markdown wiki on your disk, that an LLM gardens for you between conversations.\n\nMindBase implements Andrej Karpathy's [LLM-Wiki pattern](https://x.com/karpathy/status/1911080091498963196) as a product. You feed it sources (papers, articles, thoughts). The LLM reads, cross-references, flags contradictions, and writes structured wiki pages. Later, when you ask a question, the wiki already has the synthesized answer — no vector search re-derivation at query time.\n\n**Status:** Early access. Beta for feedback (2026-Q3). Data model + core loop stable; some UI features still in v1 → v2 migration.\n\n[Why MindBase](#why-mindbase)[How it works (30 seconds)](#how-it-works-30-seconds)[What makes it different](#what-makes-it-different)[Prerequisites](#prerequisites)[Install — choose your AI editor](#install--choose-your-ai-editor)[First project (2 minutes)](#first-project-2-minutes)[Core workflows](#core-workflows)[Working with multiple projects](#working-with-multiple-projects)[Web UI (optional companion)](#web-ui-optional-companion)[Feature matrix by editor](#feature-matrix-by-editor)[Where your data lives](#where-your-data-lives)[Architecture at a glance](#architecture-at-a-glance)[Troubleshooting](#troubleshooting)[Roadmap](#roadmap)[Beta feedback](#beta-feedback)[License](#license)\n\nYou read a lot. Papers, articles, tweets, docs. You want to remember them, connect them, form opinions from them. Today you have two bad options:\n\n**Notion / Obsidian / Roam:** Passive containers. You do all the organizing. AI features are bolted-on generation, not maintenance.**NotebookLM / Perplexity Pages / ChatGPT search:** RAG-based. Nothing accumulates. Every question re-derives the answer from raw sources.\n\n**MindBase is the third option:** the LLM actively maintains a persistent, structured wiki as you feed it sources. Knowledge compounds. Your `context.md`\n\ngets sharper every time you contribute. The AI *remembers you across sessions* because your beliefs are written down in markdown files — not stored in a chat history that gets summarized away.\n\nThink of it as **a personal Wikipedia that an AI intern writes for you**, kept up to date, cross-referenced, and honest about what it doesn't know.\n\nThree physical layers on disk (Karpathy's model):\n\n```\nLayer 1 — sources/          You add these. Append-only. LLM reads, never rewrites.\n   ├─ contributors/          Your thoughts, dated per-user (like a lab notebook)\n   ├─ research/              Wiki pages the LLM writes (concepts, entities, syntheses)\n   └─ raw/                   PDFs, URL captures, pastes — original binaries\n\nLayer 2 — README.md          The rules of your wiki. You edit. LLM reads at every op.\n          context.md         The synthesized current truth. LLM writes. You read.\n          index.yaml         Auto-generated file catalog. LLM maintains.\n\nLayer 3 — logs/              Chronological log of every operation. LLM appends.\n          artifacts/         Generated outputs: daily briefs, exports, lint reports.\n```\n\nThree operations the AI can run:\n\n| Operation | What happens |\n|---|---|\nContribute |\nYou feed a thought / PDF / URL. LLM reads, discusses key takeaways with you, updates 5-15 relevant wiki pages, appends log. |\nAsk |\nYou ask a question. LLM reads your `context.md` first (already-synthesized truth), then drills into cited source pages. Cited answers with `[[wikilink]]` provenance. |\nLint |\nLLM audits the whole wiki for contradictions, stale claims, orphan pages, missing cross-references, gaps that need investigation. |\n\nThey form a loop — good answers and lint findings flow back in as new contributions, so the wiki compounds:\n\n``` php\nflowchart LR\n    C[\"contribute<br/><small>thought · PDF · URL</small>\"] --> S[(\"sources/<br/><small>append-only</small>\")]\n    S --> B[\"build<br/><small>LLM synthesizes</small>\"]\n    B --> W[(\"context.md<br/>research/*.md\")]\n    W --> Q[\"ask<br/><small>cited answers</small>\"]\n    W --> L[\"lint<br/><small>contradictions · orphans · gaps</small>\"]\n    Q -. \"good answers filed back\" .-> C\n    L -. \"follow-up notes\" .-> C\n```\n\nThat's the entire product. Everything else is UX.\n\n**vs. Notion:** You own the data (markdown files, not a proprietary DB). The AI is a **maintainer**, not a bolt-on generator. `context.md`\n\nevolves; Notion pages don't.\n\n**vs. Obsidian:** You get the same local-first markdown vault, but with an AI that **actively writes** the wiki — you don't have to hand-craft every note. Comes with `[[wikilinks]]`\n\ncross-referencing done for you.\n\n**vs. NotebookLM / Perplexity:** Knowledge **compounds**. Ask \"what's my view on RAG?\" and the answer already exists in `context.md`\n\n. NotebookLM re-derives it from raw sources every time — nothing accumulates.\n\n**vs. any RAG tool:** RAG is stateless retrieval. MindBase is **stateful synthesis**. The wiki gets richer with every ingest. Answers get faster and more consistent over time.\n\n**Node.js 20+**(that's it for Cursor / Windsurf / Cline / Continue — the MCP server installs itself via`npx`\n\n)- One of: Claude Code, Cursor, Windsurf, Cline, Continue.dev, or another MCP-compatible AI editor\n**pnpm 10+** only if you install the Claude Code plugin or the web UI from source- (Optional) A modern browser, if you want the web UI\n\nMindBase runs **entirely on your machine**. No cloud upload, no signup, no telemetry.\n\nMindBase ships as an **MCP (Model Context Protocol) server** — published on npm as [ mindbase-mcp](https://www.npmjs.com/package/mindbase-mcp). Every MCP-compatible AI editor can use its tools with a one-line config; no clone, no build. Claude Code gets an extra layer of polish (slash commands, sub-agents) via the plugin.\n\nGet the full Karpathy 8-step ingest with sub-agents, slash commands, auto-context, and per-agent tool boundaries.\n\n**Install the plugin** — two commands inside Claude Code, no clone, no build:\n\n```\n/plugin marketplace add frankchu91/mindbase\n/plugin install mb@mindbase\n```\n\nRestart Claude Code when prompted. (The plugin launches its MCP server via `npx -y mindbase-mcp`\n\n, so npm delivers the server automatically.)\n\n## Developing the plugin from a clone?\n\n```\ngit clone https://github.com/frankchu91/mindbase.git\ncd mindbase && pnpm install\nclaude --plugin-dir apps/plugin\n```\n\n**Verify:** Open Claude Code, type `/`\n\n. You should see `/mb:contribute`\n\n, `/mb:ask`\n\n, `/mb:build`\n\n, `/mb:status`\n\n, and 8 others. Run `/mcp`\n\n— you should see `mb`\n\nconnected with 49 tools.\n\n**What you get:**\n\n- 12 slash commands (\n`/mb:*`\n\n) - 5 sub-agents with security boundaries (\n`contributor`\n\n,`builder`\n\n,`curator`\n\n,`researcher`\n\n,`migrator`\n\n) — each has a strict tool allowlist - SessionStart hook — every new Claude Code session auto-injects your project's README + context + recent log entries\n- The full Karpathy 8-step ingest: read → discuss takeaways → propose plan → wait for approval → execute\n\nSkip to [First project](#first-project-2-minutes).\n\nAdd to `~/.cursor/mcp.json`\n\n(create the file if missing):\n\n```\n{\n  \"mcpServers\": {\n    \"mindbase\": {\n      \"command\": \"npx\",\n      \"args\": [\"-y\", \"mindbase-mcp\"]\n    }\n  }\n}\n```\n\nRestart Cursor. Open Cursor Chat → the tool picker should list `mindbase_contribute`\n\n, `mindbase_ask_wiki`\n\n, and 47 others.\n\n**Optional but recommended:** teach Cursor's LLM about MindBase conventions by adding to `~/.cursor/rules.md`\n\n(or per-project `.cursorrules`\n\n):\n\n```\n# MindBase conventions\n\nWhen the user mentions \"mindbase\", \"mb\", \"my wiki\", \"记一下\", or \"记忆\",\nthe following MCP tools are available: mindbase_contribute, mindbase_ask_wiki,\nmindbase_status, mindbase_load_project, and more.\n\nRules:\n1. When user says \"add to mindbase X\" or \"记一下 X\", ALWAYS call mindbase_contribute\n   with text=X. Never just acknowledge without calling the tool.\n2. When user says \"ask mindbase X\" or \"问问我的 wiki\", call mindbase_ask_wiki\n   with query=X.\n3. When ingesting a PDF/URL, follow the 8-step Karpathy loop:\n   read → discuss 3 key takeaways with user → propose plan → wait for approval\n   → then call mindbase_contribute with the summary.\n4. Default project comes from config.json currentProjectId. If the user says\n   \"for project X\" or \"在 X 项目里\", pass projectId=X to the tool call.\n```\n\nSkip to [First project](#first-project-2-minutes).\n\nWindsurf uses the same MCP config format. Open Cascade settings → MCP → add:\n\n```\n{\n  \"mcpServers\": {\n    \"mindbase\": {\n      \"command\": \"npx\",\n      \"args\": [\"-y\", \"mindbase-mcp\"]\n    }\n  }\n}\n```\n\nRestart Windsurf. Same rules-file approach as Cursor works — put a `MindBase conventions`\n\nblock in your Cascade rules.\n\nSkip to [First project](#first-project-2-minutes).\n\nOpen VSCode → Cline extension settings → **MCP Servers**. Add:\n\n```\n{\n  \"mcpServers\": {\n    \"mindbase\": {\n      \"command\": \"npx\",\n      \"args\": [\"-y\", \"mindbase-mcp\"]\n    }\n  }\n}\n```\n\nCline auto-detects. Every tool call gets a confirmation dialog by default (transparency win). Skip to [First project](#first-project-2-minutes).\n\nOpen `~/.continue/config.json`\n\n. Add under `experimental.modelContextProtocolServers`\n\n:\n\n```\n{\n  \"experimental\": {\n    \"modelContextProtocolServers\": [\n      {\n        \"transport\": {\n          \"type\": \"stdio\",\n          \"command\": \"npx\",\n          \"args\": [\"-y\", \"mindbase-mcp\"]\n        }\n      }\n    ]\n  }\n}\n```\n\nRestart VSCode. In Continue chat, MCP tools appear under `@`\n\n. Skip to [First project](#first-project-2-minutes).\n\nIf your editor supports MCP (Zed, Aider, Goose, ChatGPT Desktop's future MCP support, custom Claude Agent SDK apps), the pattern is identical: point it at\n\n```\nnpx -y mindbase-mcp\n```\n\nas a stdio server. See your client's MCP docs for the exact config location.\n\nOnce your editor has the MCP server connected, create your first project.\n\n```\n/mb:init my-research --mission \"Study transformer attention mechanisms\"\n```\n\nClaude will scaffold `~/mindbase-data/projects/my-research/`\n\nwith `README.md`\n\n, `context.md`\n\n, `index.yaml`\n\n, and empty `sources/`\n\n, `logs/`\n\n, `artifacts/`\n\ndirectories, and set it as your current project.\n\nType in chat:\n\n```\nCreate a new mindbase project called my-research about transformer \nattention mechanisms.\n```\n\nThe LLM will call `mindbase_init_project({ name: \"my-research\", mission: \"...\" })`\n\nand confirm.\n\n```\n/mb:status              # in Claude Code\n```\n\nOr ask any other IDE:\n\n```\nwhat's the status of my mindbase?\n```\n\nYou should see:\n\n```\nMindBase — Project: my-research\nLocation: ~/mindbase-data/projects/my-research\n\nREADME     28 lines\ncontext.md 8 lines  (last built: never)\nindex.yaml 5 lines\n\nContributors: 0 users, 0 entries\nSources:      0 research, 0 raw\nLogs:         1 day\nArtifacts:    0 outputs\n```\n\nYou're ready.\n\nThe four things you'll do every day.\n\nAdd a short observation, a decision, or a hot take. Goes straight into today's contributor file, no ceremony.\n\n**Claude Code:**\n\n```\n/mb:contribute \"Attention is basically a soft dictionary lookup — each query \nretrieves a weighted average of values based on similarity to keys.\"\n```\n\n**Any other IDE:**\n\n```\nAdd to my mindbase: Attention is basically a soft dictionary lookup — each \nquery retrieves a weighted average of values based on similarity to keys.\n```\n\n**Result:** appends to `~/mindbase-data/projects/my-research/sources/contributors/<you>/2026-07-11.md`\n\nand logs the operation.\n\nFor a real source, MindBase walks through Karpathy's 8-step ingest.\n\n**Claude Code (flagship — with sub-agent):**\n\n```\n/mb:contribute /Downloads/attention-is-all-you-need.pdf\n```\n\nThe `contributor`\n\nsub-agent will:\n\n- Read the PDF fully\n- Come back to you with 3 key takeaways\n- Propose which wiki pages to create/update\n- Wait for your approval\n- Execute — creating\n`sources/research/attention-mechanism.md`\n\n, updating`context.md`\n\n, appending log - Confirm:\n`✓ Created 4, updated 3, logged.`\n\n**Any other IDE (lite mode — no sub-agent):**\n\n```\nRead this paper [/Downloads/paper.pdf] and ingest it into my mindbase. \nWalk me through the 3 key takeaways before committing anything.\n```\n\nYou get most of the value but the sub-agent's structured approval flow is a Claude Code exclusive. The LLM in Cursor/Windsurf/Cline will typically call `mindbase_contribute`\n\ndirectly — you can steer the ritual manually via your prompt.\n\nQuery with cited answers.\n\n**Claude Code:**\n\n```\n/mb:ask \"how did I resolve the attention scaling factor question?\"\n```\n\n**Any other IDE:**\n\n```\nSearch my mindbase — how did I resolve the attention scaling factor question?\n```\n\nThe LLM calls `mindbase_ask_wiki`\n\n(graph-aware retrieval: top hits + wikilink neighbors) then synthesizes a cited answer:\n\n```\nYou resolved this on 2026-06-15 [context.md:42-51]. The 1/√d_k scaling \nprevents dot products from growing too large in high dimensions, which \nwould push softmax into low-gradient regions [attention-mechanism.md:23-28].\n\nSources:\n- [[context.md]]\n- [[attention-mechanism.md]]\n- [[sources/research/scaled-attention.md]]\n```\n\nAfter a few contributions, regenerate the curated `context.md`\n\nso it reflects your current thinking.\n\n**Claude Code (flagship — with sub-agent):**\n\n```\n/mb:build\n```\n\nThe `builder`\n\nsub-agent:\n\n- Validates project structure\n- Gathers all unbuilt sources since last build\n- Synthesizes them into a fresh\n`context.md`\n\n- Writes atomically (snapshots the old one first so you can rollback)\n- Rebuilds the index\n\n**Any other IDE:**\n\n```\nRebuild my mindbase context.md — synthesize all unbuilt sources.\n```\n\nThe LLM will orchestrate the equivalent MCP tool calls. Slightly less structured than the sub-agent flow but functionally equivalent.\n\nAsk MindBase to audit itself.\n\n**Claude Code:**\n\n```\n/mb:lint\n```\n\n**Any other IDE:**\n\n```\nRun a health check on my mindbase — find contradictions, orphans, gaps.\n```\n\nOutput looks like:\n\n```\nFound:\n- 1 contradiction: attention-mechanism.md:15 says \"scaling by √d\" but \n  context.md:32 says \"scaling by 1/√d\". Same claim, different notation.\n- 2 orphan pages: [[legacy-transformer-note]] and [[old-benchmark]] \n  have no inbound wikilinks.\n- 1 gap: \"positional encoding\" is mentioned 4 times across the wiki \n  but has no dedicated page.\n- 3 suggested cross-links: [[scaled-attention]] ↔ [[transformer-architecture]] \n  (co-mentioned in 3 sources).\n\nWant to file follow-up notes for any of these?\n```\n\nYou can have as many projects as you want (`ai-research`\n\n, `insurance`\n\n, `dissertation`\n\n). One is the \"current\" project — everything defaults to it. To switch, or to route a single operation to a specific project without switching, use flags.\n\n**Claude Code:**\n\n```\n/mb:load ai-research\n```\n\n**Any other IDE:**\n\n```\nSwitch my mindbase to the ai-research project.\n```\n\nWrites `{\"currentProjectId\": \"ai-research\"}`\n\nto `~/mindbase-data/config.json`\n\n. Persists across sessions.\n\nEvery command that takes a project supports `-p <project-id>`\n\n(or `--project <project-id>`\n\n):\n\n```\n/mb:contribute -p ai-agents \"GPT-5 rumored 2026-Q4\"\n/mb:build -p insurance\n/mb:ask -p ai-agents \"my stance on agent marketplaces\"\n/mb:lint -p dissertation\n/mb:daily-brief -p ai-agents --date 2026-07-05\n```\n\nThe `-p`\n\nflag **does not change your current project**. It only routes this one operation.\n\nJust mention the project in natural language:\n\n```\nAdd \"GPT-5 rumored 2026-Q4\" to my mindbase's ai-agents project.\n```\n\nThe LLM passes `projectId: \"ai-agents\"`\n\nto the MCP tool call. Current project unchanged.\n\nMindBase includes a browser-based viewer at `http://localhost:4321`\n\n. It's not required — the plugin/MCP path is the primary interface — but useful for **browsing** what the LLM has written.\n\n`Cmd+I`\n\nfrom anywhere — a thought goes straight into today's contributor file. Dark mode included.\n\n```\npnpm --filter @mindbase/server dev\n# then open http://localhost:4321\n```\n\nYou'll see:\n\n**LeftRail tree:** Contributors, Research, Logs, Artifacts, README, Context, Raw — organized as a file tree**Article view:** click any file to read it; edit inlinequick capture that appends to today's contributor file (works in any browser tab)`Cmd+I`\n\nAddEntry:**Search (** keyword search across your wiki`Cmd+K`\n\n):**Project switcher:** dropdown at the top\n\nWhat the Web UI **can't** do (yet):\n\n- Run\n`/mb:ask`\n\n,`/mb:build`\n\n,`/mb:lint`\n\n,`/mb:research`\n\n— these need an LLM in the loop, which lives in your AI editor - The Rebuild button in the Web UI just tells you to run\n`/mb:build`\n\nin Claude Code\n\nThink of the Web UI as **a Finder for your wiki**, and your AI editor as **where you talk to it**.\n\n| Feature | Claude Code | Cursor / Windsurf / Cline / Continue | Web UI |\n|---|---|---|---|\nSlash commands (`/mb:*` ) |\n✅ | ❌ (use natural language) | ❌ |\n| MCP tools directly | ✅ | ✅ | ❌ |\n| Karpathy 8-step ingest with sub-agents | ✅ | ❌ | |\n| Auto-context via SessionStart hook | ✅ | ✅ (Cursor supports MCP hooks) | N/A |\n| Contribute short thought | ✅ | ✅ | ✅ (Cmd+I) |\n| Ingest PDF / URL / paste | ✅ | ✅ | ❌ |\n| Ask wiki with cited answers | ✅ | ✅ | ❌ (planned) |\nBuild (regenerate `context.md` ) |\n✅ | ✅ | |\n| Health check / lint | ✅ | ✅ | ❌ (planned) |\n| Deep research (web + wiki) | ✅ | ✅ | ❌ |\n| Wiki tree browsing | ❌ | ❌ | ✅ |\n| Rich file editor | ❌ | ❌ | ✅ |\n| Cross-project search | ✅ | ✅ | |\n`-p project-id` routing flag |\n✅ | N/A |\n\n**Legend:** ✅ works ·\n\nEverything is under `~/mindbase-data/`\n\n(override with `MINDBASE_DATA_DIR`\n\nenv var).\n\n```\n~/mindbase-data/\n├── config.json                       # { currentProjectId, ... }\n└── projects/\n    └── my-research/\n        ├── README.md                 # Ops manual — you edit, LLM reads\n        ├── context.md                # Synthesized truth — LLM writes, you read\n        ├── index.yaml                # Auto-generated catalog\n        ├── sources/\n        │   ├── contributors/         # Your dated entries (append-only)\n        │   │   └── <you>/\n        │   │       └── 2026-07-11.md\n        │   ├── research/             # LLM-authored wiki pages\n        │   │   └── attention-mechanism.md\n        │   └── raw/                  # PDFs, HTML captures, binary sources\n        │       └── 2026-07-11/\n        │           └── paper.pdf\n        ├── logs/                     # Chronological operation log\n        │   └── 2026-07-11.md\n        ├── artifacts/                # Generated outputs\n        │   ├── briefs/               # Daily briefs\n        │   ├── exports/              # Bundle exports\n        │   └── attachments/          # Images pasted into notes\n        └── state/                    # LLM working state\n            └── builder/snapshots/    # context.md snapshots for rollback\n```\n\n**Everything is markdown.** Open in vim, VSCode, Obsidian — all work.**No proprietary database.** No SQLite of secret sauce. What you see on disk is what MindBase knows.**Git-friendly.**`cd ~/mindbase-data && git init`\n\n— every change auditable and rollbackable.**Delete freely.**`rm -rf ~/mindbase-data/projects/foo`\n\n— nothing else needs cleanup.\n\nBackups: if you use iCloud / Time Machine / Dropbox, they'll pick this up automatically since it's just files.\n\n```\n       ┌────────────────────────────────────────────┐\n       │  Your AI editor (any MCP-compatible):      │\n       │  Claude Code · Cursor · Windsurf · Cline · │\n       │  Continue.dev · Zed · Aider · ...          │\n       └───────────────────┬────────────────────────┘\n                           │  MCP (stdio JSON-RPC)\n                           ▼\n       ┌────────────────────────────────────────────┐\n       │       MindBase MCP server (Node.js)        │\n       │       — 49 tools: contribute, ask_wiki,    │\n       │         build, lint, research, ...         │\n       │       — reads/writes ~/mindbase-data/      │\n       └───────────────────┬────────────────────────┘\n                           │  file I/O\n                           ▼\n       ┌────────────────────────────────────────────┐\n       │        ~/mindbase-data/projects/…          │\n       │        Your markdown wiki on disk          │\n       └───────────────────┬────────────────────────┘\n                           │  read-only http\n                           ▼  (optional)\n       ┌────────────────────────────────────────────┐\n       │  Web UI at localhost:4321 (React + Vite)   │\n       │  Tree browser, editor, Cmd+I add-entry     │\n       └────────────────────────────────────────────┘\n```\n\n— TypeScript strict library; storage adapters, LLM adapters, wiki index, graph, compile pipeline`packages/core`\n\n— MCP server (JSON-RPC over stdio); tools + prompts; framework-agnostic`apps/mcp`\n\n— Express server; serves the Web UI +`apps/server`\n\n`/api/tree/*`\n\nREST endpoints— React 19 + Zustand + Tailwind frontend`apps/web`\n\n— Claude Code plugin bundle (slash commands + sub-agents + hooks + bundled MCP server)`apps/plugin`\n\nCause: malformed config JSON, or the server binary can't start.\n\nFix:\n\n- Run it manually:\n`npx -y mindbase-mcp < /dev/null`\n\n— should print`[mindbase-mcp] connected · dataDir=…`\n\non stderr and exit 0. - If that works, the issue is in your editor's config (typo, wrong file, trailing comma).\n- Claude Code plugin users: verify the bundle exists —\n`ls /path/to/mindbase/apps/plugin/mcp-server/dist/cli.js`\n\n— and rebuild with`pnpm --filter @mindbase/plugin build`\n\nif missing. - Restart your editor after fixing config.\n\nCause: You haven't run `/mb:init`\n\nyet, or `config.json`\n\nis missing `currentProjectId`\n\n.\n\nFix:\n\n```\n/mb:init my-first-project\n```\n\nOr manually: `echo '{\"currentProjectId\": \"my-project\"}' > ~/mindbase-data/config.json`\n\nCause: you have a legacy v1 project directory (from before the 2026-06 refactor).\n\nFix (Claude Code only):\n\n```\n/mb:migrate --project <legacy-project-name>\n```\n\nOr delete it and start fresh: `rm -rf ~/mindbase-data/projects/<legacy-project-name>`\n\nthen `/mb:init`\n\n.\n\nCause: MCP handshake didn't complete.\n\nFix: Restart your editor. Check that no other MCP server is failing (a crash in one can block others in some clients).\n\nCause: the LLM didn't recognize the intent as a tool call opportunity.\n\nFix: Add the `.cursorrules`\n\nblock from the [Cursor](#cursor) section. Being explicit (\"call mindbase_contribute with text=...\") works too.\n\nCause: a pre-existing server-side route gap. Non-blocking — some tab count badges show 0 until fixed.\n\nFix: known issue, coming in the next patch release. Track it on the [issues page](https://github.com/frankchu91/mindbase/issues).\n\nOpen a GitHub issue: [github.com/frankchu91/mindbase/issues](https://github.com/frankchu91/mindbase/issues)\n\n**Now (2026-Q3 beta):**\n\n- Multi-editor MCP support ✅\n- Karpathy 8-step ingest via sub-agents (Claude Code) ✅\n- Cross-project routing (\n`-p`\n\nflag) ✅ - Web UI tree browser + Cmd+I add-entry ✅\n- Wiki v2 layout ✅\n\n**Next (Q4):**\n\n- Web UI\n`Ask`\n\npanel (skip Claude Code roundtrip for queries) - Web UI\n`Lint`\n\ncards (visual health check inbox) - Audio input via Whisper (\n`/mb:contribute --audio recording.m4a`\n\n) - Browser extension for one-click \"save this page to MindBase\"\n- Anthropic marketplace listing (\n`/plugin install mindbase@mindbase-official`\n\n) `mindbase_command`\n\nunified meta-tool (slash-like UX in Cursor/Windsurf)\n\n**Later:**\n\n- Team brain (multi-contributor projects with human-in-the-loop review)\n- Audio Overview (NotebookLM-style podcast digests)\n- Marp slide-deck answers\n- Tauri desktop app (one-click install for non-technical users)\n- Mobile quick-capture apps\n\nI'm running MindBase as a private beta through 2026-Q4. If you're using it:\n\n**Bug reports / feature requests:**[github.com/frankchu91/mindbase/issues](https://github.com/frankchu91/mindbase/issues)** Direct feedback / 30-min interview:**[haobing0304@gmail.com](mailto:haobing0304@gmail.com)\n\nI read every issue and reply within 24 hours during beta. If you tried MindBase and gave up — please tell me why. The blockers you hit are gold.\n\n**What kind of feedback is most useful?**\n\n- What happened the first minute after install?\n- Which command / workflow do you use every day, if any?\n- Which command / workflow did you try once and never again?\n- What did you expect to work that didn't?\n- What broke your trust in the LLM's synthesis?\n- Would you pay for this? What's the shape of the pricing that would feel fair?\n\n[MIT](/frankchu91/mindbase/blob/main/LICENSE) — do what you want, no warranty. If you build something interesting on top, I'd love to hear about it.\n\n**Built with the belief that AI's most valuable gift is not \"generation on demand\" but \"gardening of a persistent artifact you own.\"**", "url": "https://wpnews.pro/news/show-hn-mindbase-an-llm-that-maintains-a-wiki-from-your-notes-and-papers", "canonical_source": "https://github.com/frankchu91/mindbase", "published_at": "2026-07-22 04:27:11+00:00", "updated_at": "2026-07-22 04:52:14.558635+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-tools", "ai-research"], "entities": ["MindBase", "Andrej Karpathy", "Notion", "Obsidian", "Roam", "NotebookLM", "Perplexity Pages", "ChatGPT"], "alternates": {"html": "https://wpnews.pro/news/show-hn-mindbase-an-llm-that-maintains-a-wiki-from-your-notes-and-papers", "markdown": "https://wpnews.pro/news/show-hn-mindbase-an-llm-that-maintains-a-wiki-from-your-notes-and-papers.md", "text": "https://wpnews.pro/news/show-hn-mindbase-an-llm-that-maintains-a-wiki-from-your-notes-and-papers.txt", "jsonld": "https://wpnews.pro/news/show-hn-mindbase-an-llm-that-maintains-a-wiki-from-your-notes-and-papers.jsonld"}}