Local RAG for code and documentation — in VS Code
Version française
Kimono gives your MCP assistant a local search index for the project you are
working on. It combines keyword and semantic search, returns relevant file
locations first, and lets the assistant read only the passages it needs.
Use it to find code, APIs, configuration and documentation across a project
without asking an assistant to load whole files up front. Indexing and
embeddings run locally.
In the current Copilot benchmark using Gemini 3.8 Flash, Kimono RAG measured:
| Task |
Cumulative input tokens |
Elapsed time |
| Repository research |
323,368 → 105,223 (−67.5%) |
111.212 → 79.077 s (−28.9%) |
Prepare a maxDepth plan |
454,796 → 130,244 (−71.4%) |
175.224 → 84.034 s (−52.0%) |
The comparison is standard Copilot search versus local Kimono RAG. See the
benchmark results.
What you get
- Local RAG: index and search your source code and documentation on your
machine.
- Lexical and semantic search: find exact terms or related concepts.
- Office documents: search Excel (
.xlsx) cells, Word (.docx) paragraphs
and PowerPoint (.pptx) slide text and tables, with slide numbers.
- File catalogue: when content cannot be read, search the filename, project-relative
path, extension, size and modification date where available.
- Targeted context: receive file paths and line ranges before requesting
source text.
- Read-only project access: Kimono does not change your source files.
- Useful defaults:
.gitignore rules apply and common dependency, build
and cache folders are excluded.
- Included runtime: the extension bundles the server, Node.js, search
engines and embedding model. No separate runtime installation is required.
- Progress and settings: use the Kimono panel to monitor indexing and
configure the project corpus.
PDF text is searchable when pdftotext (Poppler) is installed on the extension
host. Without it, PDF files remain searchable through their filesystem metadata.
Install from the Marketplace
- Install Kimono RAG MCP from the
Visual Studio Marketplace.
- Open and trust your project folder.
- Run MCP: List Servers, start Kimono, and enable its tools in your
MCP chat client's tool picker.
- Ask your assistant to find relevant code or documentation. It can read
selected line ranges when it needs details.
Requires VS Code 1.106 or later and an MCP-capable chat client. VS Code-based
IDEs can also use the extension when they support Visual Studio Marketplace
extensions; availability depends on each IDE's extension and Marketplace
configuration.
Select a project or mode
The default mode is rag; the open workspace folder is the default project:
| Setting |
Default |
Change it when |
kimono.corpusMode |
rag |
Select aikido only to search an AI-KIDO knowledge base. |
kimono.root |
Open workspace folder |
Index a different folder. |
kimono.mode |
embedded |
Select remote to connect to an existing HTTP MCP server, or disabled to stop Kimono. |
kimono.url |
Empty |
Set the HTTPS /mcp endpoint when using remote. |
kimono.configFile |
Empty |
Supply an optional absolute path to a host-managed corpus configuration. |
Use Kimono: Open Management Panel for corpus settings, index status and a
search test. The local RAG mode does not require a special project structure.
The optional AI-KIDO knowledge mode requires prompts/index.md or index.md.
Install manually
Download a platform VSIX from
GitHub Releases,
then run Extensions: Install from VSIX… in VS Code:
| Extension host |
VSIX target |
| macOS Apple Silicon |
darwin-arm64 |
| Linux ARM64 (glibc) |
linux-arm64 |
| Linux Intel/AMD 64-bit (glibc) |
linux-x64 |
| Windows Intel/AMD 64-bit |
win32-x64 |
Choose the platform of the extension host. For Remote SSH, WSL or Dev
Containers, install on the remote host.
Use outside a VS Code-based IDE
For a shared MCP server, run Kimono with Docker Compose or the published Docker
image. For a local stdio server, build Kimono and the native search engines
from source; see the local RAG setup guide.
The published container can also provide a remote HTTP MCP endpoint for GitHub
Copilot cloud agent when deployed on a host reachable over HTTPS. Direct
integrations designed for specific LLM command-line tools are planned.
Kimono can also search an AI-KIDO knowledge base as an optional mode. Learn
about the wider AI-KIDO method at www.ai-kido.fr.
