Skip to content

manual-config.md

Latest commit

Β 

History

History
235 lines (162 loc) Β· 9.07 KB

manual-config.md

File metadata and controls

235 lines (162 loc) Β· 9.07 KB
Raw

Manual Configuration Guide

🌐 Language / 语言: English Β· δΈ­ζ–‡

Prefer to configure everything by hand? This page walks through every step. If you'd rather use the interactive wizard, use the upstream setup wizard instead.

Step 1 β€” Create your own repo from this template

Prefer Use this template β†’ Create a new repository on the GitHub page. That gives you your own copy with all the automation included, without the extra Actions friction that forks often add.

If you intentionally need an upstream-linked development copy, you can still use Fork, but this manual guide assumes you are creating your own standalone digest repo.

Optional: one-click deploy via the Setup Wizard

The recommended one-click path no longer uses a PAT. Instead:

  1. install the Linnet Bridge GitHub App on your target repo
  2. open the upstream setup wizard at https://yuyangxueed.github.io/Linnet/setup/
  3. click Install GitHub App / Authorize GitHub
  4. deploy from Step 5

This manual guide remains the fallback path when you want to keep every config change explicit, or when org / repo policy blocks the GitHub App flow.

Step 2 β€” Add your API key

In your repo go to: Settings β†’ Secrets and variables β†’ Actions β†’ New repository secret

Name Value
OPENROUTER_API_KEY Your key from openrouter.ai/keys β€” free tier works, starts with sk-or-...

This is the default fast-path credential. OpenRouter lets you call many AI models (Gemini, GPT, Claude) with one key and switch between them any time.

πŸ’° Cost estimate: With the default model (google/gemini-2.5-flash-lite), one full daily digest run costs roughly $0.1 USD. At one run per day, that's under $3 USD / month. Actual spend varies with the sources you enable, how many papers get fetched, and the summary language β€” you can watch per-call costs live on the OpenRouter dashboard. If you want to spend less, swap scoring_model / summarization_model in config/sources.yaml for a cheaper model, or lower the daily paper cap in config/extensions/arxiv.yaml.

If you prefer a different OpenAI-compatible provider, set these fields in config/sources.yaml:

llm:
  provider: "openai"
  base_url: "https://api.openai.com/v1"
  api_key_env: "OPENAI_API_KEY"
  scoring_model: "gpt-5-mini"
  summarization_model: "gpt-5-mini"

Then export or store the matching secret name instead:

export OPENAI_API_KEY=sk-...

Step 3 β€” Enable GitHub Pages

Go to: Settings β†’ Pages β†’ Source: GitHub Actions

Click Save. Your site URL will appear there β€” it looks like https://YOUR-USERNAME.github.io/Linnet.

Step 4 β€” Pick your research topics

Open config/extensions/arxiv.yaml. It has four ready-made profiles β€” uncomment the one closest to your work and edit the keywords freely:

# PROFILE A: AI / ML / LLM (general)
# categories: [cs.AI, cs.LG, cs.CL, cs.CV, stat.ML]
# must_include:
#   - large language model
#   - foundation model

# PROFILE B: Astrophysics / Space Science
# PROFILE C: Chemistry / Materials Science
# PROFILE D: Computational Biology / Bioinformatics

Want summaries in a different language? Open config/sources.yaml and change language: "en" to "zh", "fr", "de", "ja", "ko", "es", or any other language code.

Step 5 β€” Run it for the first time

If GitHub Actions / workflows are currently disabled in the repo, enable them manually first unless you already used the Setup Wizard’s Step 5 auto-enable option successfully. This is most common on forks.

Clean start for older forks: delete any inherited generated snapshots under docs/data/daily/*.json, docs/data/weekly/*.json, and docs/data/monthly/*.json, but keep the .gitkeep files. The next Daily Digest run will create fresh data for your repo.

You need to manually trigger two workflows in order:

  1. Generate the digest content: Go to Actions β†’ Daily Digest β†’ Run workflow β†’ Run workflow and wait for it to finish (about 3–5 minutes). This calls the LLM, builds today's digest, and commits it into docs/.
  2. Deploy the site to GitHub Pages: Go to Actions β†’ Deploy Astro Site to GitHub Pages β†’ Run workflow β†’ Run workflow and wait for it to finish (about 1–2 minutes). This builds the Astro site and publishes it.

πŸ’‘ From then on, Daily Digest runs automatically every day and auto-triggers the deploy on success β€” so you only need to click these two buttons the first time.

Once both are green, your site is live at https://<your-username>.github.io/<repo-name>/.

Optional but recommended: paste that URL into the repository's About -> Website field so the site link also appears on the GitHub repo home page.

Turn sources on and off

Open config/sources.yaml and set enabled: true or enabled: false for each source:

arxiv:
  enabled: true          # arXiv papers β€” the main event

hacker_news:
  enabled: true          # top Hacker News stories

github_trending:
  enabled: true          # today's trending GitHub repos
  max_repos: 15

weather:
  enabled: true
  city: "Edinburgh"      # change to your city

postdoc_jobs:
  enabled: false         # academic job listings β€” turn on if you want these

supervisor_updates:
  enabled: false         # professor/lab page monitor β€” turn on if you want these

quote_of_day:
  enabled: false         # daily quote as briefing tagline (English, requires API_NINJAS_KEY)

hitokoto:
  enabled: false         # 一言 daily quote as briefing tagline (Chinese, no key needed)

You can also switch AI models here, set llm.provider, point llm.base_url at another OpenAI-compatible provider, change llm.api_key_env, or cap how many papers get fetched per day.

Customise LLM prompts

Every summarisation and scoring prompt can be overridden in config/sources.yaml under the llm.prompts: block. The commented-out defaults are already in that file β€” uncomment and edit any you want to change:

llm:
  summarization_model: "google/gemini-2.5-flash-lite"
  # prompts:
  #   arxiv_summary: |
  #     Summarize the core method and contribution of the following paper
  #     {lang}, in 2-3 sentences (≀100 words):
  #     Title: {title}
  #     Abstract: {abstract}
  #   hacker_news_summary: |
  #     Summarize the core content of the following tech news story
  #     {lang}, in one sentence (≀50 words):
  #     Title: {title}
  #     URL: {url}

Available placeholders per prompt:

Prompt key Placeholders
arxiv_score {title}, {abstract}
arxiv_summary {title}, {abstract}, {lang}
hacker_news_summary {title}, {url}, {lang}
github_summary {full_name}, {description}, {lang}

Get your digest in ServerChan

If you want a lighter notification path, especially for Chinese-language workflows, ServerChan is a good fit:

  1. Open sct.ftqq.com/sendkey
  2. Copy your SendKey
  3. Add it as a secret: Settings β†’ Secrets and variables β†’ Actions β†’ New repository secret, name it SERVERCHAN_SENDKEY
  4. Enable it in config/sources.yaml:
sinks:
  serverchan:
    enabled: true
    max_papers: 5
    max_hn: 3
    max_github: 3
    max_jobs: 3

This keeps the key out of YAML and out of version control.

Get your digest in Slack

In addition to the website, you can receive a daily Slack message:

  1. Go to api.slack.com/apps β†’ Create New App β†’ From scratch
  2. Left sidebar β†’ Features β†’ Incoming Webhooks β†’ toggle On
  3. Scroll down β†’ Add New Webhook to Workspace β†’ choose your channel β†’ Allow
  4. Copy the webhook URL (looks like https://hooks.slack.com/services/T.../B.../...)
  5. Add it as a secret: Settings β†’ Secrets and variables β†’ Actions β†’ New repository secret, name it SLACK_WEBHOOK_URL
  6. Enable it in config/sources.yaml:
sinks:
  slack:
    enabled: true
    max_papers: 5    # how many papers to include
    max_hn: 3        # how many HN stories to include
    max_github: 3    # how many trending repos to include

If you skip this step, nothing breaks β€” the website still updates as normal.

What runs automatically

When What happens
Weekdays at 09:30 UTC by default Full digest β€” papers, HN, GitHub trending, weather, US pre-market stocks, any extras you enabled
Every Monday at 00:00 UTC Weekly summary of the past week
1st of every month at 00:00 UTC Monthly overview

The setup wizard can generate a custom daily schedule in .github/workflows/daily.yml from your local time and UTC offset. You can also edit the cron lines manually, or trigger any workflow by hand: Actions -> [workflow name] -> Run workflow.