Skip to content

Clew 0.2.0 F2

Choose a tag to compare

@miuuyy miuuyy released this 11 Jul 23:28

Generate a learning map. Click any topic. Follow the thread.

0.2.0 takes Clew from a graph experiment to a product. Three big additions: Obsidian bridge in both directions, an MCP server so external assistants can read your graph, and a real light/dark theme split.

Product positioning

  • Renamed the product to Clew.
  • New brand mark, favicon set, launch loader, and product language.
  • Reframed around personal adaptive learning paths instead of generic graph visualization.

Interface and themes

  • Full light and dark theme support (Paper light + Midnight dark).
  • Rebuilt large parts of the workspace shell. Cleaner graph controls, denser readability, smoother idle motion, stable curved edges.
  • Theme CSS split into core, overrides, workspace chat, dialog/canvas, and token files.

Obsidian bridge

  • Import: vault folder → graph package, with preview + validation. Notes become topics, wiki links become edges, folders become zones, missing links become placeholder topics, note bodies survive as artifacts.
  • Export: any Clew graph → Obsidian-ready folder with topic notes, resources, artifacts, zones, metadata, optional progress.

MCP

  • New Clew Study Assist. A read-only MCP stdio server.
  • Tools: list_graphs, get_current_learning_context, get_node, search_notes.
  • Setup docs for Claude Desktop, Claude Code, Cursor.
  • Read-only by design. Graph edits still go through Clew's propose / review / apply flow.

Local runtime and architecture

  • .env.example for faster first install.
  • Refined planner prompts and proposal validation.
  • Backend route dependencies and helpers split out.
  • Repository storage and config concerns separated.
  • Local debug log capture for frontend / API / backend errors.
  • New tests across chat persistence, MCP tools, providers, proposals, repository, quiz routes, graph canvas core, and graph interactions.

Why this release matters

You can start from a goal, notes, or messy topics; let AI draft the structure; inspect the graph; study through it; and keep the path portable through Obsidian and MCP.

Still local-first, still hackable.

image image