Curious Scan

Deal Radar

v0.2.0 · MIT

Keyword watcher for deal communities. Watches public deal feeds for the words you care about — free, promo code, giveaway — and prints matches as they appear. No API key, no account.

Install

$ git clone https://github.com/curiousscan/deal-radar
$ cd deal-radar  # nothing to install — stdlib only

Usage

# one scan, right now
$ python deal_radar.py --once --keywords free,giveaway,"promo code"

# keep watching every 5 minutes
$ python deal_radar.py --watch --interval 300 --subreddits deals,buildapcsales,GameDeals

# machine-readable output
$ python deal_radar.py --once --json > hits.json

# Claude writes a human analyst-style digest (needs ANTHROPIC_API_KEY)
$ export ANTHROPIC_API_KEY="sk-ant-..."
$ python deal_radar.py --once --claude --keywords free,giveaway,"promo code"

Options

Flag	Default	What it does
`--subreddits`	deals, buildapcsales, FreeEBOOKS, GameDeals, frugalmalefashion	comma-separated list to watch
`--keywords`	free, giveaway, promo code, coupon, 100% off	comma-separated keywords
`--once`	—	single scan pass, then exit
`--watch`	—	poll continuously until Ctrl+C
`--interval`	300	seconds between passes in watch mode
`--json`	—	emit hits as JSON
`--claude`	—	Claude-written digest of the hits (needs `ANTHROPIC_API_KEY`)
`--claude-model`	claude-sonnet-5-5	Claude model used for `--claude`

How it works

Polls subreddits' public RSS feeds — no API key, no account needed.
Matches keywords case-insensitively against post titles and bodies.
Dedupes by post ID within a run; --watch keeps polling forever.

Requirements: Python 3.8+. Stdlib only.

View source on GitHub →

Free-Model Router

v0.2.0 · MIT

One key, many free models. Send a prompt to the free tiers of several LLM providers through a single CLI — with automatic fallback when a model is busy, rate-limited, or down.

Setup (one time)

# 1. Get a free API key — no credit card required
#    https://openrouter.ai/keys

# 2. Export it
$ export OPENROUTER_API_KEY="sk-or-..."

Usage

# ask the default model chain (falls back automatically)
$ python router.py "Explain recursion like I'm five"

# prefer a specific model, fall back to the rest on failure
$ python router.py "Write a haiku about GPUs" --model google/gemma-3-27b-it:free

# pipe input, add a system prompt
$ echo "Summarize: ..." | python router.py --system "You are a terse assistant."

# see the configured free-tier lineup
$ python router.py --list-models

# benchmark the free-tier answer against Claude, side by side
$ export ANTHROPIC_API_KEY="sk-ant-..."
$ python router.py "Explain recursion like I'm five" --claude-benchmark

How it works

Talks to OpenRouter's unified chat API (/api/v1/chat/completions).
Tries your preferred model first, then walks the fallback table.
A model that errors (429/404/5xx) is skipped, not fatal.
Sends HTTP-Referer: https://curiousscan.in so free-tier usage is attributed.
Free-tier lineups rotate — check openrouter.ai/models and edit the MODELS table in router.py when a model disappears.
--claude-benchmark answers via Claude (Sonnet) too and prints both — Claude is the quality bar the free answers are judged against. Needs ANTHROPIC_API_KEY.

Requirements: Python 3.8+. Stdlib only. One free OpenRouter key.

View source on GitHub →

Trending-Repo Scout

v0.2.0 · MIT

Daily digest of the fastest-rising AI repos on GitHub. Queries the public GitHub API, filters for AI signal, renders Markdown. No API key needed.

Install

$ git clone https://github.com/curiousscan/trending-repo-scout
$ cd trending-repo-scout  # nothing to install — stdlib only

Usage

# AI repos trending over the last 7 days
$ python scout.py

# tighter window, higher bar
$ python scout.py --days 3 --min-stars 100

# save a markdown digest (great for a cron job)
$ python scout.py --out digest.md

# machine-readable
$ python scout.py --json > repos.json

# Claude (Haiku) rates each repo REAL vs HYPE — signal-vs-hype filter
$ export ANTHROPIC_API_KEY="sk-ant-..."
$ python scout.py --claude-filter

How it works

Queries GitHub's public Search API for repos created in the last N days, sorted by stars — a solid proxy for "trending".
Filters to AI-related repos via keyword matching on name, description, and topics (--no-filter disables this).
Renders a Markdown digest with stars, language, and descriptions.
Unauthenticated GitHub API allows 60 requests/hour — plenty for a daily digest.
--claude-filter asks Claude Haiku to rate every repo REAL (working code, real utility) or HYPE (marketing fluff, star-bait). Verdicts are labeled in the digest, never silently dropped. Needs ANTHROPIC_API_KEY.

Requirements: Python 3.8+. Stdlib only.

View source on GitHub →

Changelog

Release history for all three tools. Full per-release notes live on each repo's GitHub releases page.

v0.2.0 October 2026 · Claude integration

Deal Radar — new --claude flag: Claude writes a human analyst-style digest of the hits, grouped by theme, flagging the genuinely free finds. Key-gated via ANTHROPIC_API_KEY; raw hits still print if the call fails.
Free-Model Router — new --claude-benchmark flag: answers your prompt through the free-tier chain and Claude (Sonnet) side by side, so free-tier quality is judged against the Claude bar.
Trending-Repo Scout — new --claude-filter flag: Claude Haiku rates every repo REAL vs HYPE for signal-vs-hype. Verdicts labeled in the digest, never silently dropped.
All Claude features are optional, stdlib-only (direct Messages API calls, no SDK), and fail soft.

v0.1.0 October 2026 · initial public release

Deal Radar — public launch: subreddit RSS keyword watching, one-shot and watch modes, JSON output.
Free-Model Router — public launch: single CLI across free LLM tiers via OpenRouter, automatic fallback table, --list-models.
Trending-Repo Scout — public launch: GitHub Search API digest, AI-signal filtering, Markdown and JSON output.
All three: MIT license, zero dependencies, Python 3.8+.

Roadmap (from the repos): Telegram/Discord alert sink and price-drop tracking for Deal Radar; Threads/Instagram watchers; expanded model lineups for Free-Model Router.