Launched this week
Semlith
Local search over your code and documents for AI agents.
8 followers
Local search over your code and documents for AI agents.
8 followers
Coding agents make up quotes from PDFs and Office files they cannot open. Semlith is a free, open-source local index of your code and docs that Claude Code, Cursor and Codex search over MCP. In my test, made-up quotes went from 41 of 127 to 0.
Free
Launch Team / Built With
Dropbox Sign APIBuild eSignatures into your app with less work.
Promoted
Hi Product Hunt, I'm Aakash. I've been building Semlith on my own since July.
The coding agents I use kept quoting PDFs and Word files they couldn't actually open. So I tested it. I asked an Opus agent 127 questions about 53 public documents with only Grep and Read, and 41 of its quotes weren't anywhere in the file it named. With Semlith installed it found the right passage 98% of the time and made up none.
Semlith is free and open source. It's a single Rust binary that indexes your code and documents on your machine, and Claude Code, Cursor, Codex and other agents search it over MCP instead of grepping. No API keys, and nothing leaves your machine when you search. One command installs it and registers it with the agents it finds.
It doesn't help everywhere. On single-repo code with a strong model it made no difference, 92% with or without it.
If your agents also run in the cloud or in CI, or you want one index for a whole team, Semlith Cloud is the paid, hosted version. The trial is 14 days with no card, and you don't need it if all your agents run on your laptop.
What do your agents get wrong when they search your code or docs?
The made-up-quote test is the part that stuck with me — 41 of 127 to zero is a strong result. How does the index handle files that change after indexing, and when an agent gets a passage back, does it also get a file and line reference it can check, or just the text itself?
@jcbuilds Thanks. A daemon watches your folders, so when you save a file it only re-embeds what actually changed and it's searchable again in like a second. Every result comes back with the file path and line range, plus the function or heading it sits in, so the agent can open that exact span and check it. And when the agent reads that span, Semlith checks the file first, so if it changed since indexing it gets the current lines from disk, not the old copy.