Skip to main content

Exactly the right code context for your coding agent

Fewer tokens. Better recall. Zero cloud.

Coding agents burn their context window rediscovering your codebase — grep after grep, whole files read for one relevant function. Shaktiman indexes it once with tree-sitter and serves search, symbols, dependencies, and budget-fitted context over MCP, entirely on your machine.

Get StartedView on GitHub
Terminal demo: indexing a repository, hybrid search, symbol dependencies, and budget-fitted context assembly

Works with

Claude CodeCursorZedAny MCP client

Tree-sitter indexing

Functions, classes, symbols, imports, and call graphs — parsed, not regexed. Nine languages supported out of the box.

Hybrid search

Keyword (FTS5), semantic (vector), structural, and change signals combined into a single ranked result.

Budget-fitted context

Ask for 4K tokens, get exactly 4K tokens of the most relevant code — no overflow, no padding.

Local-first, always fresh

Runs entirely on your machine; your code never leaves it. A file watcher re-indexes on every save.

Up and running in a minute

git clone https://github.com/midhunkrishna/shaktiman.git
cd shaktiman
go build -tags "sqlite_fts5 sqlite bruteforce hnsw" -o shaktiman ./cmd/shaktiman
go build -tags "sqlite_fts5 sqlite bruteforce hnsw" -o shaktimand ./cmd/shaktimand
shaktiman init . && shaktiman index .

Then wire it into Claude Code via .mcp.json — see Claude Code setup for the full walkthrough.

Give your agent a better map

Get Started

MIT-licensed, open source, and local-first.