michraz

Israeli public tenders (מכרזים) in Python. One Tender object, whichever site it came from.

Sources today: Israel Land Authority land tenders, central-government tenders (mr.gov.il), Beer Sheva, and 12 more municipalities (Akko, Arad, Beit Shemesh, Dimona, Kfar Saba, Kiryat Motzkin, Nesher, Netivot, Nof HaGalil, Savyon, Tamra, Yavne). The first search takes around 20 seconds because mr.gov.il is read page by page. After that the responses are cached for an hour.

Unofficial client. It can lag or miss tenders, and parsing can break when a source changes. Confirm deadlines on the publisher's page before you act. Not affiliated with the Israel Land Authority or any municipality.

Edit in draw.io

Install

pip install michraz            # Python 3.10+
pip install "michraz[pandas]"  # optional, for DataFrames

Use it

from michraz import Michraz, to_csv

with Michraz() as m:
    tenders = m.search(keyword="גינון", open_only=True)

for t in tenders:
    print(f"{t.deadline:%Y-%m-%d}  {t.title}  {t.url}")

to_csv(tenders, "tenders.csv")

Results are sorted by deadline. Filter by source, closing_before and closing_after (dates or aware datetimes) too. Each tender has an attribution string. Keep it when you republish the data.

Some city sites list staff vacancies next to their tenders. They are hidden by default. Pass kind=None to see both, or --kind all on the command line.

Or from the terminal:

michraz search --open --source rmi
michraz search --keyword גינון --csv gardening.csv

A record that fails to parse is skipped and listed in m.last_errors. If one source is down, the others still answer and its error is listed there too. SourceUnavailableError is raised only when every searched source fails.

More scripts are in examples.

Add a source

Subclass BaseSource. Say how to read the records and how to turn one into a Tender. Skipping bad records, reporting them in errors and dropping repeats is shared.

from michraz.sources import BaseSource


class MySource(BaseSource[dict]):
    name = "my-city"
    description = "My city's tenders"
    attribution = "My City, my-city.example"

    def records(self):
        yield from self._client().get_json(self.name, "https://my-city.example/api/tenders")

    def parse(self, record):
        return Tender(source=self.name, source_id=record["id"], title=record["title"], attribution=self.attribution)

Then Michraz([MySource()]). A city that uses the shared /bids/ page needs only a name, a publisher and a domain (see michraz/sources/municipal.py).

Edit in draw.io

Fields are documented in docs/model.md, and how each source was checked in docs/sources.md.

Polite by default

Edit in draw.io

Honest User-Agent, one request per second per host, one-hour disk cache, at most two retries. Tune it with Michraz(cache_dir=None, cache_ttl=..., min_interval=...).

Limitations

• RMI region, type, purpose and status are raw numeric codes in tender.extra. No verified label table exists.
• RMI tender details and documents are not fetched yet.
• Keyword search is a Unicode-normalized substring match. No stemming.

Development

make venv && make check
make diagrams   # re-export the PNGs from docs/diagrams (needs draw.io desktop and Graphviz)

See CONTRIBUTING.md.

MIT licensed. The data belongs to its publishers and carries their licenses.