Skip to what the platform does

Democracy Routes

01 / 12 · Overview

Democracy Routes

Turn a real problem into an organised, repeatable deliberation.

Democracy Routes is a working platform for structured collective work: design the process, decide who meets whom, run the rounds online or in the room, and keep the transcript, summary and outcome attached to the issue that started it.

block types
18block types17 placeable in the builder
API endpoints
251API endpointsroute handlers in the app
services in production
17services in productionapp, rooms, transcription, workers, proxy
unit tests
500+unit testson the core logic in src/lib
02 · Why this exists

Most collective discussion is useful in the moment and gone by the next one.

The energy is real; the continuity is not. A call ends, the recording is never opened, notes sit in someone's document, and the next session restarts the same argument from zero.

The format is improvised

Who speaks, for how long, in what order — decided live, differently every time, so quality depends on whoever happens to facilitate.

The record dissolves

Recordings go unwatched and notes lose their context. Weeks later nobody can say why a decision was taken, or by whom.

The next step is unowned

A discussion ends without producing the thing that would let the next one start further along.

03 · The core ideaShipped

A problem becomes a round table; round tables compose into a route.

This is the platform's central mechanic, and the reason it is not a meeting tool. An issue enters once and keeps its history as the work moves through sessions.

  1. A problem is opened

    Someone states an issue in the platform — it exists before any meeting does.

  2. A process is designed

    A template defines the rounds, prompts, timing and grouping rules the session will use.

  3. The round runs

    Participants meet in small groups, online or in person, with a facilitator view and a control room.

  4. The record is produced

    Transcript, speaker labels, summary and participation balance are attached to the meeting.

  5. The route continues

    The output becomes the starting point of the next round instead of a file nobody reopens.

04 · Design the processShipped

Build the deliberation before it happens, then run it exactly as designed.

The flow builder is where a session stops being improvised. Blocks are composed into an ordered flow with its own timing, then launched — the runtime walks participants through it step by step.

  • Grouping

    Form rooms at random, by opposing stance, or to mix positions deliberately.

  • Discussion

    A timed small-group round in an embedded video room.

  • Prompt

    Put a question, poster or brief in front of everyone at the same moment.

  • Notes

    Private written reflection captured against the block that produced it.

  • Form

    Structured answers and choices, aggregated across participants.

  • Solo recording

    One private spoken contribution per participant.

  • In-person table

    A physical table with its own recorder, for rooms rather than calls.

  • Table move

    Regroup the room between rounds so people meet new tables.

  • Embed

    Bring approved external material into the flow.

  • Pause / Break

    Deliberate silence and rest, timed like everything else.

Eleven interaction blocks ship today. Partner modules for external deliberation tools are in progress and are not presented here as working integrations.

05 · Who meets whomShipped

The platform decides who should talk to whom — and changes it every round.

Grouping is not shuffling. Rooms can be formed at random, split across opposing positions, or mixed to break echo chambers, and the algorithm avoids re-pairing people who have already met.

  • Random

    A neutral baseline when the topic does not call for anything sharper.

  • Polar

    Pairs participants across a stance divide so disagreement is actually met.

  • Anti-repeat

    Prefers partners nobody has met yet, so a long route keeps widening contact.

The pairing logic is pure, unit-tested code (src/lib/services/flowMatching.ts). The animation runs the same repeat-avoidance rule.

06 · Core workspaceShipped

Community spaces are the long-term container for collaboration.

A community space keeps related people, meetings, templates, texts and analytics together, so a group builds a shared working memory instead of scattering the process across isolated pages.

  • Members

    People connected to the same space, with shared visibility and collaboration rights.

  • Meetings

    Live or in-person sessions tied to the same ongoing context.

  • Templates

    Reusable deliberation structures that can be reopened, edited and launched again.

  • Texts

    Asynchronous written material that complements meetings between sessions.

  • Analytics

    Space-level summaries and recurring analysis across meetings and templates.

07 · Meeting lifecycleShipped

A meeting produces a record, not just a call.

Live rooms, transcription, summaries, speaker labels and participation balance all become part of the same meeting record — during the session and after it ends.

  • Embedded live room

    The call runs inside the app, with participant context and facilitator controls.

  • Transcription

    Live or post-call, across several providers depending on the meeting mode.

  • Speaker labels

    Generic diarised speakers are mapped back to the real participants.

  • Participation balance

    Speaking time per participant, as a post-call reflection aid.

08 · Entry pointShipped

The issue can exist before the meeting does.

Someone publishes an open problem, connects it to a community space, finds whether a similar issue already exists, and gathers people around it before deciding what session should follow.

  • Similar problems

    Before opening a new issue, check whether a related one is already running.

  • Space connection

    A problem can belong to a community space, or wait until the right context appears.

  • Join before meeting

    People join the issue first, then decide together what the next round should be.

09 · AI supportShipped

AI is support infrastructure, not an invisible decider.

Transcription help, summaries, analysis and drafting assistance — each tied back to the meeting evidence that produced it, so the output can be checked rather than trusted blindly.

  • Transcription

    Several providers, including a browser-based worker that runs Whisper locally.

  • Summaries

    Post-call recaps and readable outputs generated from the meeting transcript.

  • Analysis

    Meeting- and space-level analysis across templates, meetings and texts.

  • Assistants

    Drafting help and configurable in-meeting agents that support, not replace, facilitation.

10 · Where it fitsShipped

One workflow, several very different rooms.

Civic groups, institutions, researchers and organisations enter for different reasons and share the same need: stronger continuity from one discussion to the next.

  • Civic groups

    Recurring discussion around local issues, campaigns and shared texts.

  • Institutions

    Better-structured input than a comment form, with an inspectable process.

  • Research

    Study discussion formats, participation balance and how routes evolve.

  • Organisations

    Team problem-solving, design sessions and cross-functional coordination.

11 · Under the hoodShipped

Already running as a coordinated multi-service stack.

Not a mockup: the live deployment combines the main app, embedded video rooms, transcription services, workers, event tracing and analysis.

  • dr-appControl planeUsers, community spaces, meetings, templates, texts, open problems, admin.
  • dr-videoLive roomEmbedded WebRTC room with recording, live transcription and media tracing.
  • transcription-hubTranscript storeNormalised transcript sessions and deliberation artifacts.
  • dr-event-hubEvent layerStructured runtime events for debugging and operational visibility.
  • dr-remote-workerWorker runtimeBrowser-based post-call transcription with job claiming.
  • audio-apiSpeech to textSelf-hosted Deepgram and Vosk transcription paths.
  • dr-thinkerAnalysisMeeting and flow analysis surfaces used by the main app.
12 · Outcome

Built so every serious discussion leaves something behind.

Context, evidence, process memory and a route that can keep growing — the basis for the next round rather than a recording nobody opens.