Kenya's parliamentary record, open and queryable.

What is Bunge Hub?

Bunge Hub is an open-source platform that exposes Kenya's parliamentary record as structured, queryable data. Every bill debated, every contribution made, every question raised, presented as it actually happened, directly from the Hansard, with no editorial layer in between.

You can look up any Member of Parliament and read exactly what they said on any bill or topic. You can follow a bill's journey through Parliament stage by stage and see who spoke at each point. You can search across thousands of sittings, questions, and debates in one place.

The platform currently covers the 13th Parliament of Kenya (September 2022 to present). The codebase and data pipeline are openly available for anyone to run, extend, or build on.

Where does the data come from?

The raw transcripts are fetched and parsed by odnelazm, an open-source scraper and data pipeline built by Mwananchi Tech. It reads parliamentary sittings from mzalendo.com, extracts bills, members, topics, and contributions into a structured database, and runs the AI enrichment that generates summaries.

Mzalendo itself is maintained by the Mzalendo Trust, a Kenyan non-profit that aggregates data from the official Kenya Hansard, the verbatim record of everything said in Parliament, published by the National Assembly and the Senate. Member profiles, photos, party affiliations, and committee memberships also come from Mzalendo's member performance tracker.

Full credit and gratitude to the Mzalendo Trust for making this data publicly available, and to the Parliament of Kenya for publishing the official Hansard record that makes all of this possible.

How does it work?

Bunge Hub uses an automated program to regularly read and process the transcripts published on Mzalendo. For each parliamentary sitting, it reads the full record of what was discussed, identifies bills being debated, picks out the names of members who spoke, and organises everything so it can be searched and browsed.

Think of it as a thorough research assistant who reads every Hansard document, highlights the key parts, and organises them into a library you can explore. The search works even if you spell a name slightly differently or only remember part of a bill's title.

AI summaries

Summaries of bill debates, topics, and sittings are generated by open-source AI models reading the full transcript. Each summary displays a small label showing the model used. The full contribution text is always available, and the original transcript is the authoritative record.

Pages not yet processed show an "AI enrichment pending" notice.

Known limitations

Because the platform reads transcripts automatically rather than having a human review every page, there are a few things to be aware of:

• Speaker names may vary. The Hansard records names exactly as they appear in the transcript, which sometimes differs between sittings due to typos or different formatting. The same member may appear under slightly different names across records.
• Bill identification is not perfect. Bills are detected from section headings in the transcript. Occasionally a heading that resembles a bill title may not actually be one, or a genuine bill debate may be listed under a heading that does not match its formal name.
• Legislative stages may be missing. When a bill is debated without a clear stage label in the heading, the stage is left blank. This is common for committee debates and some procedural items.
• Presiding officers show no activity. The Speaker and Deputy Speakers are recorded in transcripts by role label rather than by name, so their contributions cannot be reliably linked to their member profiles.
• AI summaries may be imprecise. The model reads the full transcript but may miss nuance, misattribute emphasis, or produce a summary that does not fully reflect the complexity of a debate. Always read the original transcript for anything consequential.
• Coverage is not exhaustive. Only sittings indexed by Mzalendo are included. If a transcript has not been published or is incomplete there, it will not appear here.

If you spot something wrong, please let us know on GitHub. We will do our best to address it.

Open source

Bunge Hub is open source and freely available. The code is on GitHub under the GPL-3.0 licence. Contributions, bug reports, and suggestions are welcome.

• mwananchi-tech/bunge-hub