Markpilot — Code Notes to Obsidian & AI
Install from the VS Code Marketplace · 中文文档
Code notes & highlights synced to an Obsidian vault, BYOK multi-provider LLM Q&A with long-term memory, and streaming selection translation. The core pipeline is shared with the Markpilot Chrome extension (web notes / page Q&A), so your memory and profile carry over between browser and editor.
Features
- Add Note on Selection (editor context menu) — attach a note to any selection; the selection gets a yellow highlight, and the note's line anchor shifts automatically as you edit the code above it.
- Ask AI about Selection — sends the selection to the sidebar Chat as context; questions are answered with the current file, your notes on the file, and long-term memory (when a vault is configured) injected.
- Translate Selection — streams a translation of the selection into the sidebar (target language follows your VS Code display language), with one-click Replace Selection.
- Export File Notes to Obsidian — writes this file's notes to
<vaultPath>/Markpilot-Code/code-<file>.md (idempotent, full-file rewrite matched by frontmatter source).
- Long-term memory — Q&A is augmented with retrieval results from
<vaultPath>/Markpilot-Memory/ (same hybrid recall as the Chrome version); a _profile.md user profile, if present, is injected too.
- Bring Your Own Key — 9 provider presets (OpenCode free models, OpenAI-compatible, Ollama, OpenRouter, Anthropic, DeepSeek, Zhipu GLM, Moonshot Kimi, Qwen). Keys are stored in VS Code SecretStorage, never in
settings.json.
Quick start
Zero-config path — no API key needed:
- Command Palette → Markpilot: 快速配置 (Quick Setup)
- Pick opencode as the provider, then a free model such as
mimo-v2.5-free
- Select some code → right-click → Ask AI about Selection
For fine-tuning, use Markpilot: 打开设置 (Open Settings — gear icon in the Chat view title): a full-page settings UI with provider picker, API key status, live model list fetching, and a connection test. Changes apply immediately.
Settings
All settings live under the markpilot. prefix (the settings page edits them for you):
| Key |
Default |
Description |
provider |
openai-compatible |
opencode / openai-compatible / ollama / openrouter / anthropic / deepseek / zhipu / moonshot / qwen |
model |
"" |
Model ID; falls back to the provider's first preset when empty |
baseUrl |
"" |
Only needed for openai-compatible |
vaultPath |
"" |
Absolute path to your Obsidian vault; memory/export disabled when empty (Q&A still works) |
memoryInject |
true |
Inject long-term memory into questions |
exportAiQA |
false |
Include AI Q&A history when exporting notes |
semanticRecall |
off |
local = on-device MiniLM semantic recall (see below) |
Semantic recall (optional)
semanticRecall: "local" runs an on-device Xenova/all-MiniLM-L6-v2 ranker (model cached in the extension's globalStorage). @xenova/transformers is a lazily-loaded optional dependency and is not bundled in the vsix — if you want this feature, run npm i @xenova/transformers in the extension directory. When unavailable, recall silently degrades to lexical-only retrieval; everything else is unaffected.
Privacy
- API keys live in VS Code SecretStorage; the settings UI only shows whether a key exists, never the key itself.
- Notes and memory are stored locally (VS Code globalState + your own Obsidian vault). Nothing is uploaded to any Markpilot server — there is none.
- Network traffic goes only to the LLM provider you configured. See PRIVACY.md for details.
Links
