Research that
shows its work
Every run is a receipt. Scira plans its searches, reads deep, and shows you the source behind each sentence before you decide to believe it.
3 runs a day, free, no card.
Web search powered byCloudflare Workers caps CPU time, not wall clock, so a loop that mostly awaits runs far longer than the stated limit.
1Vercel Fluid bills the awaited time instead.
25M+
Searches answered
100K+
People researching
—
GitHub stars
100+
Connected apps
What a run actually does
Four steps, all of them visible
Nothing happens off-screen. You can open any step and see the queries, the pages, and the reasoning that produced the sentence above it.
- 01
It plans
A question becomes sub-questions. Each gets the model and the source that suit it, run in parallel where they do not depend on each other.
workers cpu time limit 2026vercel fluid max durationdeno deploy timeout+3 - 02
It reads
Eighteen built-in modes plus whatever apps you connect. Results are deduplicated by canonical URL before anything is written.
Limits · Cloudflare Workers docsdevelopers.cloudflare.com
Fluid compute · pricing and limitsvercel.com
Ask HN: what do you run long agent loops on?news.ycombinator.com
- 03
It cites
No claim without a source. Where two sources disagree it says so rather than picking the one that reads better.
Workers caps CPU, not wall clock.
1Conflict
Two sources disagree on whether the cap applies per request or per invocation. Both are shown.
- 04
It acts
Connect an app and its tools join the run. Anything that writes asks first, and says what it will change before the buttons appear.
Notioncreate_pageNeeds youWants to add “Runtime limits, Feb 2026” to the Engineering database.
Allow onceAlways allowSkip
Acts in the apps you already use
Connect a server once and its tools become part of the run, with approval on anything that writes.
What people say
Better than Grok at digging up information from X. Insanely accurate answers.
Also said
Does a really good job scraping through the reddit mines.
Searched for myself in extreme mode. Not just the best, it's wild. And 100% accurate.
Read nothing the whole sem and topped my mid-sems with Scira. Diagrams, points, topics, all there.
Questions
Full pricing details on the pricing page. Anything else, zaid@scira.ai.
- What is Scira?
- An open-source AI assistant built for research and action. It searches the web in depth, cites its sources, and connects to 100+ apps via MCP so you can act on what you find.
- Free vs Pro vs Max?
- Free gives you 3 runs/day with basic models. Pro ($15/month) gives you unlimited research, all base models, Claude Haiku unlimited, Sonnet 4.6 (150 msgs/mo), Opus 4.7 (30 msgs/mo), PDF, voice, Lookouts. Max ($60/month) adds the full Claude pool (80/wk), Gemini 3.1 Pro (80/mo), Qwen 3.7 Max (40/wk), AI image generation, and Scira Agent.
- How are model quotas enforced?
- Pro caps Claude Sonnet 4.6 at 150 messages and Opus 4.7 at 30 messages per calendar month. Max pools all Anthropic Claude requests at 80 per calendar week. Pro has a weekly pool of 100 OpenAI requests across the GPT 5.4 family and GPT 5.6 Luna and Terra. GPT 5.5 and GPT 5.6 Sol are Max-only and count toward that same pool on Max. Max also caps Qwen 3.7 Max at 40 requests per calendar week. Limits reset automatically and are visible in your account.
- Can I cancel anytime?
- Yes. Benefits stay until the end of the billing period.
- Which AI models does Scira use?
- Grok, Claude, GPT, Gemini, and others. Switch per query.
- How does Scira keep answers accurate?
- Every claim links to a retrieved source with an inline citation, and agents cross-check sources before writing.
Ask it something hard
3 runs a day, free, no card. See the sources it read before you trust the answer.