Skip to content

README.md

Latest commit

聽

History

History
65 lines (45 loc) 路 1.86 KB

README.md

File metadata and controls

65 lines (45 loc) 路 1.86 KB
Raw

Continue.dev MisakaNet Search Context Provider

This integration lets Continue users search MisakaNet lessons from inside the AI chat panel by typing @misaka <query>. It returns the lesson title, score, file path, and a relevant snippet so the model can ground its answer in failure-memory knowledge instead of asking you to leave the editor.

What it provides

  • AI-tool integration: runs inside Continue as a custom context provider.
  • Stable result shape: each item includes title, score, path, and snippet.
  • Local/offline search: uses the checked-out MisakaNet repository and the existing BM25/RRF search engine.

Install

  1. Clone MisakaNet and install its core dependency:

    git clone https://github.com/Ikalus1988/MisakaNet.git ~/MisakaNet
cd ~/MisakaNet
python3 -m pip install misakanet-core

  2. Copy config.ts into your Continue configuration, or merge the contextProviders entry into your existing ~/.continue/config.ts.

  3. If your checkout is somewhere else, update MISAKANET_REPO in config.ts.

  4. Restart Continue.

Usage

In Continue chat, type:

@misaka database locked sqlite

Continue will inject results like:

MisakaNet search results for "database locked sqlite":

1. SQLite database locked on WSL/NTFS
   Score: 0.92
   Path: lessons/contrib/sqlite-database-locked-wsl-ntfs.md
   Snippet: ... move the sqlite database to the ext4 filesystem ...

You can then ask the assistant to apply the relevant fix in your project.

Standalone JSON helper

The provider shells out to scripts/misaka_search_json.py, which is also useful for other AI tools:

cd ~/MisakaNet
python3 scripts/misaka_search_json.py "database locked" --top 3

The command prints JSON containing lesson titles, normalized scores, paths, and snippets.