Only what matters should reach your brain.
Cyris reads your RSS feeds and newsletters and hands you one digest of the articles worth reading, short enough to finish. You pick every token that reaches your model's context window; Cyris does the same for your attention.
On the maintainer's own deployment, 56 issues from 2026-08-28 to 2026-09-27: an average issue took in 92 articles and kept 13. That digest is public at cyris-digest.pages.dev, written in Traditional Chinese.
- A digest you can finish. An LLM scores every article against the topics you care about and keeps the few worth reading.
- One summary per news story, instead of the same story from five outlets.
- Nothing hidden. Every digest links to the full list of what it collected, dropped articles included.
- Votes that teach it. 👍 or 👎 any article; later runs can skip ones close to your downvotes.
- Delivered where you read: a web archive, your inbox or Discord, with an alert when a run fails.
- Any LLM, or none: Anthropic, Gemini, OpenAI, Cloudflare Workers AI, or plain excerpts with no API key.
| Local | Cloudflare | |
|---|---|---|
| Runs | cyris run on your machine, by hand or from cron |
on a schedule, in a Cloudflare Container |
| You read it | as HTML files on disk | in a web archive at a URL |
| Web votes, email-only newsletters | no | yes, with optional Workers |
| Needs | Python 3.12+ and uv | Workers Paid (US$5/month), Docker and bun; your own domain for email features |
| Guide | docs/install-local.md | docs/install-cloudflare.md |
Each guide runs top to bottom, so you can also hand it to a coding agent. The Cloudflare guide has so far been verified on one account only.
- How sources are processed: lanes, votes, and what is not supported
- Operating a deployment: updating, rolling back, checking what runs
- Hosting and cost: what one deployment spends
- Architecture: how it is built; read it before changing code
Planned, not built yet:
- Podcasts and YouTube channels as sources. You will add them like any other feed.
On a
read-lane source, cyris will read the episode itself instead of its show notes: the transcript a podcast publishes in its feed, or the video through Gemini, which needs a Gemini API key. On other lanes it will tell you a new episode is out. A setting picks the model or turns this off. A podcast that publishes no transcript will be announced but not summarized; reading its audio directly is a later step.
To change cyris, start with CONTRIBUTING.md: setup, the
scripts/check.sh gate, commits and CI. Coding agents read AGENTS.md.
AGPL-3.0-or-later © 2026 musingfox
Self-host, use, and modify cyris freely. If you run a modified version as a network service, the AGPL requires you to offer users its source. For use outside AGPL terms (e.g. a closed-source/commercial deployment), contact the author about a commercial license.