Carbide

Fork of Otterly — Carbide extends Otterly with a redesigned UI, deeper search/query workflows, language intelligence, source control, AI, and an extensible plugin/MCP platform.

Carbide is a local-first Markdown workbench for notes, documents, tasks, references, graphs, and plugin-assisted workflows. Vaults are ordinary folders on disk; Carbide adds editing, search, links, Git history, document export, AI tooling, and MCP/plugin integrations around those files.

Why "Carbide"?

Before modern flashlights, coal miners dropped calcium carbide into water to produce a bright acetylene flame in their lamps to navigate deep tunnels. The icon depicts a miner's lamp with Markdown symbols (> and *) forming the flame because Carbide is built for navigating deep personal knowledge bases with local files, search, graph views, language tools, and extensibility.

What it does

• Local-first Markdown vaults: Vaults are plain folders. Notes, attachments, queries, settings, indexes, and plugins live on disk and remain portable.
• Editing and documents: Visual/source Markdown editing, syntax highlighting, wikilinks, transclusion, outlines, PDF/document viewing, and Markdown-to-PDF export.
• HTML artifacts as first-class citizens: .html files are searchable, embeddable (![[file.html]]), and renderable in Source/Safe/Live modes with per-file trust grants — drop LLM-generated artifacts straight into your vault. See HTML Artifacts.
• Search, queries, and metadata views: Hybrid FTS + semantic search with recency-weighted omnibar, inline @ palette, composable note/task queries, search graph, and Bases views over frontmatter.
• Graph and relationships: Full-vault, active-note, and search graphs with backlinks, attachments, semantic edges, and canvas views.
• Language intelligence: LSP infrastructure for Markdown (IWE, Markdown Oxide, Marksman, rumdl) and common code languages, with diagnostics, hovers, code actions, and rename/definition.
• Tasks and planning: Markdown-native tasks with list/kanban/schedule views, hierarchical heading scoping, relative due dates, and quick-capture.
• Git-aware workflow: Init, stage, commit, checkpoint, diff, restore, remotes, push/fetch/pull/sync, optional auto-commit on save.
• References and linked sources: BibTeX/CSL/RIS, Zotero Better BibTeX integration, citation picker, bibliography export, linked PDF/HTML source folders with content-addressed extraction cache.
• AI assistance: CLI-based providers (Claude Code, Codex, Ollama), inline ask/edit with accept-reject, and vault/editor-aware prompt composition.
• Extensibility and MCP: Sandboxed plugins with marketplace, a permissioned SDK/RPC surface, Carbide's own MCP server, and plugin-hosted external MCP sidecars.

Installation

Homebrew (macOS)

brew install --cask TranscriptionFactory/tap/carbide

GitHub Releases

Download pre-built binaries from the Releases page:

A built-in auto-updater applies updates on startup.

macOS Gatekeeper

Mac users may see "Carbide is damaged and cannot be opened" because the app isn't signed with an Apple developer key. Fix it by running:

xattr -cr /Applications/carbide.app

Or open System Settings → Privacy & Security and click Open Anyway in the Security section.

Features

Markdown Editing & Documents

• ProseMirror/CodeMirror-based visual + source editing with Shiki, KaTeX, callouts, tables, and code blocks
• Wikilinks, heading/block references, backlink tracking, unresolved-link diagnostics, and rename-aware link repair
• Inline embeds for notes, headings, blocks, images, PDFs, HTML artifacts, audio/video, files, and Mermaid diagrams
• Edit-in-place for transclusions via a pencil action on the embed toolbar
• Split panes, draggable/pinned tabs, zen/focus modes, line numbers, zoom, and Vim-style navigation
• Document outline, block transforms, drag-and-drop, find/replace, copy as Markdown/HTML
• Daily notes and configurable naming templates

Document Rendering & Export

• Markdown-to-PDF via a self-contained HTML renderer + native platform PDF capture (macOS, Windows, Linux)
• PDF export inlines images, renders Mermaid diagrams, and rasterizes SVG/KaTeX math
• Mermaid SVG export and Mermaid rendering in slides export
• In-app viewers for PDF, image, code, HTML, and CSV; Excalidraw canvases; external viewer windows
• Linked PDF/HTML source folders keep source material in place with content-addressed extraction cache

HTML Artifacts

• .html files are first-class vault citizens — searchable through FTS (stripped body text + <title>), linkable as attachments, and embeddable via ![[file.html]]
• Three render modes — Source / Safe / Live — with per-file or per-folder trust grants persisted under .carbide/trusted_html.json
• Default-deny: Live mode requires an explicit grant; Live + Network is a separate opt-in
• Paste Clipboard HTML as Artifact writes the file, inserts a transclusion, and records provenance in a .meta.json sidecar
• Provenance banner above the renderer; ✕ clears the sidecar
• Sandboxed iframe envelope (allow-scripts only, no allow-same-origin); theme variables injected so artifacts can blend with the app

See HTML Artifacts for render modes, trust grants, transclusion, paste-from-clipboard, and the security model.

Search, Queries & Navigation

• Omnibar (Cmd/Ctrl+K): unified entry for notes, commands, settings, links, and cross-vault search, ranked by an exact_prefix > substring > fuzzy rule with a 24h recency boost
• Inline @ palette: insert notes, files, headings, dates, tags, citations, and commands from the editor
• Hybrid search: SQLite FTS5 + local semantic embeddings merged via Reciprocal Rank Fusion
• Structured query language: clause-based note/file/folder queries (named, with, in, linked from, with_property) with boolean composition, regex, subqueries, and saveable .query files
• Task queries: status, path, section, text, tag, and due-date filters with boolean operators, relative dates, and hierarchical heading scoping (section under <heading>)
• Fuzzy + hierarchical tag matching: #parent matches #parent/child; typos fall back to fuzzy across all tags
• Search graph and Bases (UI-driven frontmatter views — table/list/kanban/gallery/calendar)

See Search & Queries.

Tasks & Planning

• Markdown-native tasks ([ ], [-], [x]) with 3-state cycling and inline due dates
• List, kanban, and schedule views; quick capture from app/menu actions
• Embedded task query results with source-note navigation and in-place status toggles
• Hierarchical section scoping, boolean operators, relative dates, sort/group/limit

Graph, Links & Canvas

• Full-vault, active-note, and search graphs with WebGL/Pixi rendering and worker-based force layout
• Backlinks, outlinks, attachments, smart links, semantic similarity edges, with toggleable edge types
• Fuzzy filtering, cluster detection, focus mode, radial layout
• Graph-to-canvas export, Excalidraw canvases, Mermaid rendering with cached output

Language Intelligence & Tooling

• LSP results panel for diagnostics, hover, code actions, and workspace edits
• Markdown LSP via IWE, Markdown Oxide, or Marksman; rumdl-backed lint/format/fix
• Code LSP for Python, Rust, TypeScript, JavaScript, Go, JSON, and YAML when the language server is on PATH (lookups memoized)
• Managed toolchain installs Markdown tools on demand
• IWE structural transforms: extract/inline sections, list↔section, link creation

AI Assistance

• CLI-based provider presets for Claude Code, Codex, and Ollama; auto-resolution by local CLI availability
• Ask/edit modes with inline accept-reject flow
• Prompt builder composes requests from vault and editor context; AI-generated file-tree descriptions

References, Citations & Linked Sources

• Citation library (CSL-JSON) with BibTeX/CSL/RIS import/export via Citation.js
• Zotero Better BibTeX integration: connection testing, live search, import, annotation sync
• Citation picker inserts [@citekey]; bibliography rendering/export
• Linked PDF/HTML source folders with metadata extraction, search inclusion, and a content-addressed extraction cache that survives renames

See Bases & References.

Git & Source Control

• Init, status, stage/unstage, commit, commit-all, file diffs
• Named checkpoints, version history with restore, remotes, push/fetch/pull/sync
• Optional auto-commit on save

Plugins, MCP & Extensibility

• Iframe-sandboxed plugin runtime with marketplace browse/install/update/uninstall and per-plugin settings
• Permission-gated SDK for vault IO, editor, commands, UI contributions, search, metadata, network fetch, AI execution, export, and external sidecars
• Bundled plugins: Auto-Tag, HTML Strip, HTML to Markdown, PDF Export, Slides Export, Smart Templates, Wiki Compiler
• Carbide MCP server with authenticated setup flows for Claude Desktop and Claude Code; tools cover notes, search, tasks, graph, metadata, references, and Git
• External MCP sidecars via the sidecar.* SDK

See Plugin How-To.

Layout & Customization

• Sidebar views: explorer, dashboard, starred, graph, tasks, tags, source control, daily notes
• 20+ themes (light/dark), CSS token editor, rebindable hotkeys, command-palette actions
• Layout variants, context rail, outline/tasks panels, zen/focus modes, per-vault settings

More

• Embedded terminal (xterm + WebGL) with multiple sessions and respawn
• File watcher, atomic writes, asset cache, indexed metadata pipeline
• Hierarchical tags panel with prefix selection and tag-based opening
• Per-vault and global storage documented in Data Storage

Tech Stack

Documentation

• Getting Started — first-run flow, core actions, and quick links
• Search & Queries — omnibar, inline @ palette, query language, search graph, bases, task queries
• Bases & References — database views, citations, linked sources, Zotero
• Markdown Syntax Guide — supported syntax and embeds
• HTML Artifacts — render modes, trust grants, transclusion, paste-as-artifact, security model
• Architecture — decision tree, layering rules, and project map
• Plugin How-To — build plugins, use permissions, call RPC/SDK APIs, and run external MCP sidecars
• HTML to Markdown Plugin — bundled conversion plugin details
• Data Storage — where Carbide stores settings, caches, indexes, plugins, and vault data
• UI Design System — semantic tokens, components, and styling rules

Development

pnpm install
pnpm dev            # Start the Tauri dev server

Validation:

pnpm check          # Svelte/TypeScript type checking
pnpm lint           # oxlint + layering rules
pnpm test           # Vitest unit/integration tests
cd src-tauri && cargo check   # Rust type checking
pnpm format         # Prettier formatting

Acknowledgments

Carbide is a fork of Otterly with inspiration from various tools, including file management architecture from Ferrite.