StackGrep Engine: grep for your AI agents. We're not the LLM; we make yours find things.

A super fast search and indexing engine for your own documents. Bring your own data: push it over an API or point us at your S3 bucket. Your agents get exact and regex answers in milliseconds, with no embeddings to build. API docs · Quickstart · Try it on npm's source

Measured on our production servers

• 0.5 to 32 ms warm search over 100,000 documents
• 28 to 148 ms cold search
• 21,000 durable writes a second
• About 0.5 KB stored per KB of text, index included

Three calls

1. Send documents: POST /api/collections/NAME/docs, or connect an S3 bucket
2. Search: GET /api/collections/NAME/search?q=...&regex=1&filter=key:value
3. Or let your agent do it over MCP: list_collections, search_collection, count_collection, get_document

Bring your own bucket

Point a collection at a prefix in an S3 bucket you own. We read only that prefix, index every text object, and keep up as files are added, changed and deleted.

Use cases

• Hosted grep for your AI agents: Coding agents grep instead of using vector search. StackGrep gives every agent hosted grep over your documents and S3 data: exact and regex, indexed, in milliseconds.
• Exact counts for RAG and agents: Vector search returns the top k, so “how many contracts mention X?” gets a guess. StackGrep counts every matching document exactly, for your LLM or agent.
• Exact match search for AI agents: Give your AI agent a search tool over your own documents: exact text and regex, metadata filters, exact counts, answers in milliseconds. REST and MCP.
• RAG without embeddings: Retrieval-augmented generation with no embedding pipeline: your LLM finds passages by exact text and regex, with filters and counts. Nothing to re-embed when data changes.
• Grep an S3 bucket: search the contents of every file: Point StackGrep at an S3 prefix and search every object's contents by exact text or regex in milliseconds, from your code or your AI agent. Read-only access.
• A regex search API that doesn't scan everything: A hosted regex search API: run regular expressions over millions of documents in milliseconds, with metadata filters and exact match counts. No scanning everything.
• A knowledge base MCP server for your documents: A hosted MCP server that lets Claude, Cursor and Codex search your documents by exact text and regex: list, search, count and read. OAuth sign-in or API key.
• Find every contract that says exactly this: Find every contract with a given clause, term or amount, counted exactly. Exact text and regex search over your documents, with metadata filters. Built for agents.

Compared

• StackGrep Engine vs Elasticsearch
• StackGrep Engine vs Algolia
• StackGrep Engine vs Typesense
• StackGrep Engine vs Meilisearch
• StackGrep Engine vs turbopuffer
• StackGrep Engine vs Pinecone

Written in Tern

Every part of StackGrep, from the crawler and the index to the regex engine and the API, is written in Tern, a new compiled language that reads like Python and runs like C. Tern is closed source today and opens soon at ternlang.dev.

Questions