Awesome Tech Feeds

RSS, but treated like infrastructure.

A Git-native registry of technical RSS/Atom feeds with stable IDs, curated collections, deterministic builds, health probes, a CLI, and a typed Python API.

Catalog · Registry JSON · Essential OPML · SDK · Curation · Contributing

_{Geo Gecko · git pull && read better feeds.}

tl;dr

The repo currently ships 50 sources, 8 collections, and 4 profiles. The essential collection is an intentionally small 18-source starter pack.

$ techfeeds export --collection essential --format opml -o essential.opml
$ techfeeds export --topic security --trait independent --format json
$ techfeeds profile developer --registry generated/registry.json --format opml -o developer.opml

Want the raw data? Start with generated/registry.json. Want something your feed reader can import? Grab an OPML file from generated/.

Why this is a repo instead of a spreadsheet

A feed URL is easy to collect. Keeping a useful registry alive is the hard part: domains move, endpoints drift, duplicates appear, taxonomies rot, and hand-edited exports diverge.

Awesome Tech Feeds keeps the source of truth reviewable in Git and compiles consumer artifacts from it.

sources/*.yaml ───────┐
collections/*.yaml ───┼──> validate ──> compile ──> generated/registry.json
profiles/*.yaml ──────┘                         ├──> generated/*.opml
                                                └──> generated/catalog.md

scheduled probe ────────────────────────────────────> health.json release asset

The useful invariants are deliberately boring:

source.id        = durable identity
feed.url         = transport endpoint
collection       = curated bundle
profile          = executable preset
generated/       = build output
health snapshot  = network observation

That makes changes diffable, CI-checkable, and easy to consume without running a service.

Quick start

For a local checkout:

git clone https://github.com/GeoGeekLab/awesome-tech-feeds.git
cd awesome-tech-feeds
python -m venv .venv
source .venv/bin/activate
python -m pip install -e '.[dev]'

techfeeds validate
techfeeds compile --check
pytest

For the published v0.5 package:

python -m pip install \
  https://github.com/GeoGeekLab/awesome-tech-feeds/releases/download/v0.5.0/awesome_tech_feeds-0.5.0-py3-none-any.whl

Use the registry

Export exactly what you need instead of maintaining your own pile of feed URLs.

# Starter pack for a feed reader
techfeeds export --collection essential --format opml -o essential.opml

# Independent security writing as JSON
techfeeds export --topic security --trait independent --format json

# Topic filters compose with AND semantics
techfeeds export --topic ai --topic llm --trait research --format json

Filters cover collection, topic, trait, language, and source kind. Unknown values fail explicitly.

The built-in collections are:

essential · ai · systems · security · databases · programming-languages · company-engineering · independent

Every collection is ordinary YAML under collections/ and compiles to an importable OPML artifact.

Query it from Python

from techfeeds import Query, Registry

registry = Registry.from_url()
result = registry.query(
    Query(
        collection="essential",
        topics=("ai",),
        traits=("research",),
    )
)

for source in result.sources:
    print(source.id, source.website)

For pinned workflows, verify a downloaded registry before parsing it:

registry = Registry.from_file(
    "registry.json",
    expected_sha256="<digest from SHA256SUMS>",
)

The consumer surface is documented in docs/consumer-sdk.md.

Source records are code-reviewable data

A source gets a permanent ID; the website and feed URLs can change around it.

schema_version: 2
id: simon-willison
name: "Simon Willison’s Weblog"
kind: individual
language: en
website: https://simonwillison.net/
feeds:
  - url: https://simonwillison.net/atom/everything/
    format: atom
    role: primary
    official: true
topics: [ai, llm, python, databases, web]
traits: [original, practitioner, deep-dive, independent, high-frequency]
curation:
  rationale: "Admitted for recurring practitioner writing with reproducible technical detail."
  reviewer: GeoGeekLab
  reviewed_at: "2026-09-20"
  review_after: "2026-12-19"
provenance:
  added_by: GeoGeekLab
  added_at: "2026-09-20"
status: active

Field-level rules live in docs/registry.md. Curation rules live in docs/curation.md.

Profiles are presets

Profiles turn the same registry into role-oriented starting points:

developer
ai-engineer
founder
researcher

They are versioned YAML. A profile selects collections, applies explicit exclusions and boosts, and resolves deterministically.

techfeeds profile researcher --registry generated/registry.json
techfeeds profile developer --registry generated/registry.json --format opml -o developer.opml

See docs/profiles.md for the v2 profile contract.

Feed health

Network checks are kept out of the editorial source records.

techfeeds probe --concurrency 8

The scheduled workflow publishes the latest snapshot as a stable release asset:

https://github.com/GeoGeekLab/awesome-tech-feeds/releases/download/health-latest/health.json

Probe states are intentionally operational: healthy, degraded, stale, and broken. They describe what happened when the endpoint was checked; the registry remains the curated source of truth.

Generated artifacts

generated/ is committed build output, so consumers can use the project with curl, a feed reader, Python, or plain Git.

Artifact	Use it for
registry.json	Full machine-readable registry
registry.min.json	Compact client payload
all.opml	All active primary feeds
essential.opml	Small starter pack
profile-*.json	Resolved profile output
profile-*.opml	Feed-reader imports for profiles
catalog.md	Human-readable catalog

Repository map

.
├── sources/       # source records
├── collections/   # curated bundles
├── profiles/      # role presets
├── registry/      # controlled vocabularies
├── schema/        # JSON Schemas
├── src/techfeeds/ # CLI + validator + compiler + SDK + probe
├── generated/     # compiled JSON / OPML / catalog
├── tests/         # contract tests
├── docs/          # design + contracts
└── brand/         # Geo Gecko + project artwork

Send a PR

A good source PR answers three practical questions: who publishes it, why it is repeatedly useful, and where the canonical feed lives. Put that evidence in the source record; CI handles the mechanical checks.

techfeeds validate
techfeeds compile --check
pytest

Start with CONTRIBUTING.md. For larger behavior or schema changes, read docs/architecture.md first.

Scope

This repository owns source metadata, feed endpoints, curated bundles, deterministic exports, and endpoint health observations. Article fetching, summarization, ranking, delivery, and reader-specific behavior belong in downstream consumers built on top of the registry.

License

Code and tooling: MIT. Registry metadata: CC BY-SA 4.0.

---

git pull && read better feeds

Geo Gecko · Changelog · Issues