scira-cli
TUI/CLI interface where Code meets Research
Scira CLI is a command-line research tool that uses AI to answer questions by searching the web, gathering sources, verifying claims, and writing a report. You ask it a question, it plans a research approach, pulls in web results, checks the claims it finds, and saves the output as a Markdown file on your own machine. Nothing goes to a cloud service; all run data is stored locally in a folder called .scira/runs.
You install it as a Node.js package and then run a setup command that walks you through connecting it to an AI model provider and a web search provider. Several AI providers are supported, including Vercel AI Gateway, xAI (Grok), Cloudflare Workers AI, and Hugging Face. For search, you can use Exa, Firecrawl, or Parallel. Each requires an API key from the respective service. The tool has a built-in doctor command to check whether your keys are configured correctly.
You can use it in three modes. The default interactive mode opens a text-based interface in your terminal with a session history home screen. A headless mode runs the question directly and exits when done, which is useful in scripts. A classic shell mode gives a simpler line-by-line interface. You can also resume a previous run, export results as Markdown, JSON, or CSV, and set up scheduled monitoring of a topic with automatic diffing to track what changed.
Each research run saves several files: the original question, the agent's plan, incremental notes, a list of sources with URLs, a list of verified claims, and the final report. You can inspect any of these files directly since they are plain text or JSON.
Configuration is optional and handled through JSON files at the user level or per project. Settings control things like which AI model to use, how many sources to gather, whether claims must all be cited, and which domains to include or exclude from search. The project is released under the MIT license.