@blackwell-systems/knowing

Self-adapting code intelligence engine. Gives AI agents ranked, graph-aware context instead of grep results. Gets smarter with scale, not dumber.

Install and verify

npm install -g @blackwell-systems/knowing
knowing version   # should print the version

Configure your agent

Add to your agent's MCP config (.mcp.json for Claude Code, .cursor/mcp.json for Cursor, .vscode/mcp.json for VS Code, see all):

{
  "mcpServers": {
    "knowing": {
      "command": "knowing",
      "args": ["mcp", "--watch"],
      "transport": "stdio"
    }
  }
}

The MCP server auto-indexes your repo on first launch (10-30 seconds). No model downloads, no API keys required.

First useful query

Ask your agent:

"Use the context_for_task tool to find symbols related to [something you know exists in your code]."

You should see ranked symbols with scores and file paths. If results are empty, the repo is still indexing. If results seem unrelated, use specific symbol names in your task description.

What it does

knowing indexes code across 23 extractors (Go, TypeScript, Python, Rust, Java, C#, and more) into a content-addressed knowledge graph. 38 edge types, 28 MCP tools, 277 equivalence classes bridging task vocabulary to code symbols.

P@10 = 0.330 across 302 tasks, 17 repos, 8 languages. 13 self-adapting mechanisms. 3.79x codegraph, 6.00x GitNexus.

CLI usage

knowing add .                                          # index a repo
knowing context -task "refactor auth" -format gcf      # ranked context
knowing test-scope -files internal/auth/handler.go     # affected tests
knowing why -task "refactor auth" -symbol "SessionHandler"  # explain ranking
knowing enrich lsp                                     # LSP enrichment for higher-quality edges

Documentation

Full docs at https://blackwell-systems.github.io/knowing

Source: https://github.com/blackwell-systems/knowing