Core concepts

Semantic Search

Knowit works out of the box with text and tag matching. Add an OpenAI API key and search becomes embeddings-backed.

Enabling it

Set OPENAI_API_KEY in the environment where Knowit runs (your shell for the CLI, or the MCP server config for agents):

Terminal

$ export OPENAI_API_KEY=sk-...
$ knowit search "how do we handle payment retries"

With embeddings enabled, queries match on meaning rather than exact words — the query above finds an entry titled "Webhook retry policy" even though it shares almost no keywords with it.

Without a key

No key, no problem: Knowit falls back to text and tag matching. Because entries are typed, scoped, and tagged, plain matching stays useful — good titles and tags carry a lot of signal.

Never commit real API keys. Use environment variables or your MCP client's env block, and keep .env files out of version control.