|
|
|
|
Show HN: Librarium – fan out research queries to many search/AI APIs in parallel
(github.com)
|
|
1 points
by joeykudish
117 days ago
|
|
|
I wanted deep research from inside my agents. Codex, Claude Code, and other AI coding agents don't have built-in deep research. They can search the web, but they can't fan out a query to Perplexity Deep, OpenAI Deep, Exa, Brave, etc. Librarium is a CLI that does exactly that. One command, many providers in parallel, structured markdown output. Because it's a CLI, any AI agent that can run shell commands can use it. I use it as a skill in Claude Code and it works out of the box. It also deduplicates citations across providers, so you can see which sources keep showing up independently, which is a useful trust signal when you're researching something you don't already know well. |
|
|
There are three tiers of providers you can mix and match:
- deep-research: Perplexity Deep, OpenAI Deep, Gemini Deep (thorough, takes a couple minutes) - ai-grounded: Perplexity Sonar, Brave Answers, Exa (fast, comes back with citations) - raw-search: Brave Search, SearchAPI, SerpAPI, Tavily (just links, very fast)
They all run in parallel. Output is a directory with per-provider markdown files, a run.json manifest, and a sources.json that ranks citations by how many providers independently cited the same URL.
FYI about Costs: the deep-research providers aren't cheap. But you can run just the quick group (ai-grounded tier) for a fraction of the cost and it's often good enough.
To get started:
Setup API keys in your .env and then: Would love to hear what you'd use this for and what providers you think are missing. There's a request template in the repo.Example run here: https://gist.github.com/jkudish/65bf115000fb17815cba5ba42c13...