Self-host a private AI research assistant that searches the web and returns cited answers using your own API keys.
Upload a PDF and ask questions about its contents to get specific, sourced answers rather than generic summaries.
Run multi-source research across academic papers, Reddit, and GitHub simultaneously in one query.
Build a custom AI search product by forking the codebase and swapping in your preferred AI models and search backends.
| zaidmukaddam/scira | giscus/giscus | aurelia/framework | |
|---|---|---|---|
| Stars | 11,652 | 11,667 | 11,683 |
| Language | TypeScript | TypeScript | TypeScript |
| Setup difficulty | moderate | easy | moderate |
| Complexity | 3/5 | 2/5 | 3/5 |
| Audience | general | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires API keys from multiple providers, an AI model provider, Exa AI for web search, and Upstash for rate limiting, before the app is functional.
Scira is an open-source, AI-powered search and research tool that takes a question, a PDF, or a URL and produces a cited answer. Instead of returning a list of links, it breaks your question into sub-tasks, searches the web (and other sources), cross-checks the evidence, and gives back a grounded response with clickable citations so you can verify every claim yourself. The project supports 17 search modes. You can search across the general web, academic papers, Reddit discussions, GitHub repositories, YouTube videos, stock markets, cryptocurrency data, and more. There is also a chat mode for talking directly to the AI without any search, and specialized modes for things like Spotify, prediction markets, and real-time posts on X. A handful of modes (Connectors, Memory, Voice, and XQL) are marked as Pro features. Scira includes about 28 built-in tools that the AI can call on your behalf. These cover web retrieval, financial data, weather, maps, flight tracking, movie and TV lookups, code execution in a sandboxed environment, document search over uploaded files, and personal memory storage. You can upload a PDF or spreadsheet and ask questions about its contents. The codebase is built with Next.js and Tailwind CSS on the front end, the Vercel AI SDK for connecting to AI models, Exa AI for web search and content retrieval, and Upstash for rate limiting. It supports a wide range of AI model providers, including xAI, OpenAI, Anthropic, Google, Mistral, DeepSeek, and many others. Scira is AGPL-3.0 licensed, which means you can self-host it and bring your own API keys and models. A hosted version is available at scira.ai.
Scira is an open-source AI search tool that answers questions with cited sources, it searches the web, academic papers, GitHub, Reddit, and more, then returns a verified answer with clickable citations instead of just links.
Mainly TypeScript. The stack also includes TypeScript, Next.js, Tailwind CSS.
AGPL-3.0, free to self-host and modify, but any modified version you distribute or run as a network service must also be open-source under AGPL.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly general.
This repo across BitVibe Labs
Verify against the repo before relying on details.