Daily Intelligence — a personal news brief for accountants

Too much is changing to follow it all: tax rules, e-invoicing mandates, MTD deadlines, AI tools, regulator guidance. This is a small open-source worker that reads the news for you and sends you one short email a day with only the items that matter to your practice — and, for each one, why it matters to you.

You describe yourself once, in plain English:

PROFILE = {
    "who": "Sole practitioner in the UK with ~80 small business and landlord clients.",
    "jurisdictions": ["UK", "EU"],
    "topics": ["Making Tax Digital", "VAT & e-invoicing", "AI in accounting practice"],
    "watch_for": "HMRC deadline changes affecting small landlords, and whether any AI tool is actually safe to use on client data",
}

Every morning you get an email like:

MAKING TAX DIGITAL HMRC confirms MTD IT software list update — HMRC added three products to the recognised list and tightened the bridging-software criteria. If your landlord clients use spreadsheets, check their bridging tool is still listed before the next quarterly update.

Built by OpenAccountants — the open-source network of accountant-verified tax rules for AI.

How it works

RSS/Atom feeds (your jurisdictions + topics)      Claude web searches (your queries)
                    └──────────────┬──────────────────────┘
                                   ▼
                 Claude scores every candidate against YOUR profile
                 (skip / worth a read / act on this — with "why it matters to you")
                                   ▼
                 One HTML email, sectioned by your topics (via Resend)
                                   ▼
                 SQLite remembers what you've seen (no repeats for 30 days)

One run costs a few cents of Claude API usage. Your profile never leaves your own deployment: config.py is gitignored, and the only services called are Anthropic (scoring/search) and Resend (sending). No client data is involved anywhere — this reads public news only.

Setup (15 minutes)

Not technical? Paste this README into Claude or ChatGPT and say "help me set this up" — the steps below are exactly what it will walk you through.

1. Get your two API keys

• console.anthropic.com → API key + a few dollars of credit (a daily digest costs roughly $3–10/month depending on how many feeds/searches you configure).
• resend.com → API key, and verify a sending domain or address. The free tier is plenty for a personal digest.

2. Create your profile

git clone https://github.com/openaccountants/daily-intelligence.git
cd daily-intelligence
cp config.example.py config.py
# edit config.py — fill in PROFILE, timezone, and send hour
cp .env.example .env   # fill in your keys and email addresses

3. Test it locally

python3 -m venv .venv && .venv/bin/pip install -r requirements.txt
set -a; source .env; set +a
.venv/bin/python main.py --now      # runs once and emails you

4. Deploy it (Railway)

The worker just needs to stay running; Railway is the path of least resistance (~$5/month):

1. Push your clone to a private GitHub repo (your config.py stays out of git either way — see the alternative below) and create a Railway service from it. The repo includes .python-version + mise.toml so the build works out of the box; the Procfile starts python main.py.
2. Set the variables from .env.example in the Railway service.
3. Add a volume mounted at /data and set SQLITE_PATH=/data/digest.db so the "already sent" memory survives redeploys.
4. Keeping your profile out of git entirely: set DIGEST_CONFIG_B64 to base64 < config.py — the worker writes config.py from it at boot, so you can even deploy straight from this public repo with zero changes.

That's it. It sends daily at the hour/timezone in your config; if a run fails, you get an alert email that says why (including "your Anthropic credit ran out", the most common one).

Customising

• Jurisdictions & topics come from starter packs in presets.py (currently UK, US, EU, MT + eight topics). Anything not in a pack still works — an unknown topic becomes a section plus a web search automatically.
• Power users can override anything the presets produce — CONTEXT, SECTIONS, SEARCH_QUERIES, FEEDS — at the bottom of config.py.
• Models: defaults to Claude Sonnet for both search and scoring; switch SCORING_MODEL to Haiku in config.py to cut cost.

Contributing: add your jurisdiction

The most valuable contribution is a feed pack for your country: a handful of live RSS feeds for your profession's press and your tax authority, plus one or two search queries. Add them to presets.py and open a PR. Please check feeds actually return recent items before submitting — dead feeds are silently useless.

License

AGPL-3.0 © OpenAccountants. Run it, modify it, share it — if you offer a modified version as a service, share your changes back.