[pilot proposal] vTaiwan x Agora Citizen Network

[nicobao]

Context

Agora Citizen Network is a social app for public debate and deliberation, originally funded by the EU’s Next Generation Internet Initiative (NGI). Agora is designed as a fully open-source public infrastructure for collective intelligence.

Key technologies

• Bridging algorithms inspired by Pol.is, to map disagreements and surface common ground
• LLM-powered summaries, which generate meaningful group labels instead of just “Group A, B, C”
• ZK-based login (with optional phone verification) for privacy and sybil resistance.

What we propose

We propose piloting Agora for broad-listening in a way similar to how vTaiwan uses Pol.is.

The benefits include:

• More robust Sybil resistance
• Richer, AI-generated summaries that are more understandable to citizens (end users)

To enable comparison, we’ve built a feature to import past Pol.is conversations into Agora. For example:

• Original Pol.is report: https://pol.is/report/r3nhe9auvzhr36dwaytsk
• Agora conversation view: https://agoracitizen.network/feed/conversation/ss_4Cg
  In this conversation, both Polis and Agora have calculated 2 clusters although with slight difference in the number of clustered participants. Instead of group A and B, Agora has generated the labels: Optimists and Cautionists, based on the representative opinion data.

Our approach to open-source

• 100% open-source, various copyleft licenses
• Not a cathedral, not a bazaar, but an organized bazaar. Inspired by projects like ZeroMQ.
• We're not just open-source but also open-contribution, and we've already onboarded several external contributors. We're always seeking to onboard more people!

Our Next Steps

More features

• Translation, internationalization
• organization page
• [long-term] develop the Racine metaprotocol, to make Agora compatible with the various decentralized protocols such as AT Protocol, Farcaster, Waku, Nostr, Activity Pub and more, as well as with various digital identity providers and authentication mechanisms. The goal is to give agency for users to choose where they store their data and which authentication method they prefer, while providing options for highly censorship resistant methods, for self-hosted alternatives and for ways to audit otherwise centralized options
• implement decentralized moderation algorithms and UX
• Move beyond Polis-style interactions to make Agora more conversational and engaging, while still protecting users from trolls. Additionally, enhance the tool's sensemaking and collective intelligence capabilities to generate more actionable insights at the conclusion of each conversation.
• ... for more see Product Roadmap / Prioritization Matrix zkorum/product#34 (comment) (public roadmap)
• we're open to suggestions in general, as the product is always under active development

More adoption

Partner with people and organizations to carry out pilot projects that can grow into long-term engagements. We also want to support others in running their own independent instances of Agora, which is especially important in highly sensitive political contexts. For example, public consultations run by governments.

[bestian]

@nicobao

Dear Nicolas,

Thank you for sharing your team’s work. I believe that after localization into Chinese, it will certainly benefit the vTaiwan community and various deliberative democracy teams using Chinese.

May I ask if you currently need assistance with translation? How can interested contributors participate?

In addition, could you try exporting the original comments CSV data from the Agora report and paste it here or email it to me?

If the original comments CSV file already contains the fields comment-id, comment_text, votes, agrees, disagrees, and passes (note: comment-id uses a hyphen, comment_text uses an underscore), then this dataset can be directly processed by SenseMaker for analysis.

Currently, the raw CSV formats from polis.tw and pol.is have some differences in field names and other minor variations. I have written two conversion scripts to handle them. If I can review a sample of the original comments CSV exported from Agora reports, I can also write another script to accommodate any discrepancies. But if no conversion is needed and the data is directly compatible, that would of course be ideal. Thank you!

Cheers,
Bestian

p.s. Since my primary language is Traditional Chinese (Taiwan), I am not very proficient in English listening, speaking, reading, or writing. I usually draft my emails in Chinese, then use AI to translate, followed by manual proofreading. If the wording is not ideal, I kindly ask for your understanding.

[nicobao]

Hi Bestian,

I'm really glad to hear that you're interested in piloting Agora :)

We've just shipped the new i18n feature and translated the app into French and Spanish: zkorum/agora#307
We're planning to add more languages soon, and we'd love to work with you on adding Traditional Chinese!

Right now, we don't use any third-party translation software -- every translation is managed through the .i18n.ts files in the source code. For someone comfortable with Git, contributing translations is already straightforward. However, I am working on simplifying the export/import process for these files so that translators can work on them more easily.

In parallel, we're currently implementing the automatic translation of the AI-generated labels and summaries, based on user language preference.

The underlying math in Agora comes from an open-source Python library called red-dwarf, which we've sponsored the broader Polis community to develop. Since the original Polis "math" source code is written in Clojure and tightly coupled with its database, red-dwarf was created as a standalone alternative.

Some highlights of red-dwarf:

• No database dependencies -- just a simple library exposing functions
• MPLv2 licensed
• Community-led, not tied to any single organization or company
• Flexible and adaptable for multiple algorithms and AI use cases
• Designed with plurality in mind

A voluntary open-source contributor recently added an export feature to red-dwarf, which we plan to integrate into Agora. Would you be able to open an issue on the red-dwarf repo describing the requirements for compatibility with polis.tw data format? That way, we could start a discussion and explore solutions -- whether merging the approaches or simply offering a configuration variable to select the desired "export flavor".

As for AI summaries and labels in Agora, here's what we’re currently doing with a single-prompt approach:

• Generate labels for each opinion group
• Produce summaries of opinion groups based on representative opinions in the context of the conversation body/title
• Create a summary of the overall conversation based on the top common ground, divisive, and majority opinions (not displayed directly within the app anymore, it will be removed as the single-prompt approach for this task is not ideal)

This works reasonably well for labeling and summarizing opinion groups, though accuracy can suffer when conversations don't yet have many representative opinions. The conversation-level summary in particular is less accurate than what Sensemaker can achieve, since our single-prompt approach lacks the richer context that a multi-prompt process provides.

We're working on streamlining this prompt for labelling and group summary purpose only, which seems to fit the single-prompt approach.

We'd love your feedback on this and we are very interested in seeing how Sensemaker could integrate with Agora to improve the process, especially regarding the whole-conversation analysis.

We're also exploring multiple potential features:

• quote reply: let participants build on other people's opinions and creating a new one, giving them a sense of organic discussions like in most apps
• live group chats for people within the same opinion groups, with an AI learning from these conversations and then discussion with the AI from the other opinion groups, in order to help the participants iterate and build constructive proposals

P.S: on language, I totally understand -- english isn't my native language either, so please excuse any mistakes :)

[bestian]

Hi @nicobao

I just opened an issue about "Clarification on CSV Format Needed to Build a Converter for SenseMaker Integration" on Red-Dwarf's Repo.

If you have any additional information, please follow that issue here: polis-community/red-dwarf#102

[bestian]

Hi @nicobao ,

Thank you for the detailed explanation. I've completed the translations (Traditional Chinese, Simplified Chinese, and Japanese) and have filed a pull request. Please review it when you have time. If there are any structural errors or omissions, please help proofread them if convinient. Thank you.

zkorum/agora#318

As for Red-Dwarf, I'll find time to check in with them in the next few days. Thank you.

[nicobao]

Hi Bestian,

Agora is now translated in Traditional Chinese, Simplified Chinese, and Japanese.
Let me know what are the next steps for you to integrate Agora within your toolkits.
I am happy to assist in helping you setting up a local instance if you prefer the self-hosted approach, though if I understand well you first need to have the export feature ready.

[bestian]

Hi @nicobao ,

Thank you for merging the deployment. It’s great to see that live. I also shared the translated Agora Citizens site at vTaiwan’s online meeting, and everyone agreed that the interface and design look very modern and high-quality.

And thank you as well for pointing out that the export feature is really needed to connect with Sensemaker.