Research you can verify.
Code you can trust.
Two jobs, one terminal. Scira researches the web with cited sources and works in your codebase, saving every run as plain files you own.
Bring your own model and search
Grounded research
Scira searches Exa, Parallel, or Firecrawl and reads the pages before it writes a word. Every claim carries a citation you can open. No source, no claim.
Works in your code
It reads files, makes edits, greps the repo, and runs commands, and it asks first. Manual, suggest, or auto approval, so nothing touches your shell without a yes.
Yours, on disk
plan.md, sources.jsonl, claims.jsonl, report.md. Every run is plain files under .scira/runs/. Nothing is uploaded. Resume, re-run, verify, and export any time.
Nothing it does is hidden from you.
A research run moves through four stages, and each one writes a plain file the moment it happens.
plan.mdPlan
Scira breaks your question into sub-questions and a search plan you can read or edit before anything runs.
sources.jsonlGather
webSearch finds pages; readUrl pulls each one in full. Raw extracts are saved so a citation can always be re-checked.
claims.jsonlVerify
Each factual statement is tied to a source URL and a status. Claims without backing never reach the report.
report.mdReport
A written report where every sentence traces to a page you can open. Export to Markdown or JSON, or keep going in the TUI.
Configure once. Return anytime.
Pick a model and search provider from the home screen, then stay in the terminal. Every control is a slash command away.
Stop hoping your answers are true. Verify them.
MIT licensed. Runs locally. Your data never leaves your machine.