Every feature.
In plain language.

Instead of filling out a survey, you have a real conversation with an AI mediator. It asks open-ended questions, listens for what you actually care about (not just your position), and helps you put your values into words. It is instructed to understand you — never to argue or persuade. Your transcript stays in your browser; it is not stored on our servers.

Talk with the mediator in Spanish, Chinese, Vietnamese, Tagalog, Korean, Russian, Arabic, Portuguese, French or Haitian Creole as well as English. You see a summary of what was recorded in your own language; it is stored in English so everyone's perspectives can be combined.

We all have blind spots. When your wording suggests a common thinking trap — like overweighting a single bad experience, or assuming nothing should ever change — the mediator gently asks about angles you might have missed. It never tells you you're wrong and never names the "bias" to you.

Don't want to type? Tap the microphone icon and speak your thoughts. Your browser transcribes what you say. This makes participation accessible to people who find typing difficult, are on mobile, or simply prefer talking.

As you talk, a side panel shows the values and needs the system is picking up, so you can see whether it is understanding you. It is a preview — your final perspective comes from a full extraction at the end.

If you need to close the browser or step away, your conversation is saved in this browser. Come back later on the same device and pick up where you left off. Saved transcripts are deleted automatically after 30 days.

After your interview, the system would show you someone who seems to disagree with you on the surface but shares your core values. The matching engine exists; it is not yet part of the interview flow.

Individual conversations are analyzed together. The system doesn't just count votes — it groups people by their underlying values and finds what they actually agree on, what conditions would change their minds, and where they genuinely differ. Nothing is published for an issue until at least 10 people have taken part.

Combine the conditions people named ("add a trial period", "keep parking alternatives") and see an estimate of how support might change. It is a simple model built from what participants said — a way to explore options, not a forecast.

Ask questions about a published synthesis in plain English: "What do people think about parking?" "What's the biggest point of disagreement?" Answers cite specific items from the synthesis.

The dashboard shows the latest published synthesis for an issue and refreshes itself every minute, so new results appear without reloading. Real-time push updates are on the roadmap.

"What would the consensus look like if we only counted a certain group?" Explore how different subsets of participants shape the outcome. Not yet available in the product.

Every synthesis run creates a new numbered version, and older versions stay viewable, so you can see how the picture changed as more people took part. A side-by-side "track changes" view is in development.

Compare your community's discussion with patterns from similar communities and published policy precedents. Not yet available in the product.

See consensus on a map and find gaps in participation by area. Not yet available — location is not currently collected.

Map how issues relate to each other and find values a community shares across everything it cares about. Not yet available in the product.

Connect current issues to earlier debates on the same topic and track how support has shifted over time. Not yet available in the product.

Compare officials' votes with what the community said it agrees on. Not partisan — just data. In development; the accountability section currently shows empty states.

Compare where the budget goes with what the community prioritizes. In development.

Look at whether donations from particular sectors line up with votes against community consensus — correlation, not causation. In development.

Grade process quality: were hearings held, documents available before the vote, dissent acknowledged? In development.

When a council or agency acts on an issue, the decision is recorded with its source and compared, point by point, with what the community said: which concerns and conditions it addressed, partly addressed, ignored or contradicted. Recorded votes appear next to the community's stance.

A composite grade combining representation, budget, transparency, outcomes, responsiveness and equity. In development.

A per-issue timeline from promise to hearing, vote, implementation and outcome, flagging drift between what was promised and what was delivered. In development.

Track corporate commitments (e.g., affordable-unit pledges) against delivery, weighted by how enforceable they are. In development.

Estimate which officials are most likely to vote against community consensus, from factors like campaign finance and prior alignment. Research stage.

Compare how local media frames an issue with what participants actually said. In development.

A non-partisan comparison of candidates against community consensus. No endorsements. In development.

Get notified when a tracked issue reaches an agenda, a vote is recorded, or a records deadline approaches. In development.

Get a ready-to-deliver 2-minute testimony script pre-filled with your community's published consensus data. Walk into a public hearing prepared.

A petition drafted from the published consensus. Each demand cites the share of participants who raised it. Review and edit before circulating.

A starting-point public records request for documents related to the issue. Fill in your agency and details; check your jurisdiction's records law before sending.

Find organizations that share values on an issue and draft a shared platform statement. In development.

Map decision-makers, allies and targets for an issue. In development.

A ready-to-paste social post, an op-ed template, and a one-page briefing, each pre-filled from the published consensus. Copy with one click.

When a public comment period opens, start from a comment that summarizes what participants agreed on — with the number of self-selected participants stated plainly.

A starter list of volunteer tasks — testifying, canvassing, monitoring coverage — to turn findings into action. Currently a fixed list, not tailored to each issue.

Flag when a consensus item is also backed by law (e.g., ADA accessibility). In development.

Let participants distribute priorities across budget categories through the interview. In development.

At the end of your interview, one careful AI pass turns the conversation into a structured summary: your position, whether you support or oppose, the values and needs behind it, and the conditions that would change your mind — without names or identifying details.

Extracted values are also tagged with the closest types from the Schwartz model of basic human values, which helps show how different values relate to each other.

Each contribution gets a rough score for how reasoned it is — whether it gives reasons, considers others, proposes solutions. It is a simple keyword heuristic, not a judgment by the AI.

Instead of just saying "82% agree", the system shows a range — for example, "likely between 74% and 90%". It also shows a margin of error, and is honest that because people chose to take part, the real uncertainty is larger than that margin.

Your browser sketches the structure of your argument — claim, reasons, what would change your mind — and keeps it only on your device. It is not used in the synthesis.

After a decision is implemented, compare what happened with what the synthesis suggested. In development.

Submit photos, screenshots or written testimony that support, challenge or add context to claims about an issue. Attachments are images only (JPEG, PNG or WebP, up to 4 MB) — to share a document, photograph or screenshot it. Before anything is sent you review suggested redactions of personal information; images have hidden metadata (like photo GPS location) stripped; a moderator approves each item before it is public. Submissions are not verified.

After the interview you can optionally share an age band, whether you rent or own, and your area. When the organization running the issue supplies population figures, each report shows which groups are underrepresented and what the result looks like reweighted to the population. Groups of fewer than five people are never shown.

When your perspective is saved you get a receipt — a fingerprint of it. Each published synthesis lists the fingerprints of exactly the perspectives it was computed from, so you can check yours was included, right in your own browser. Anyone can recompute the published root.

One document combining consensus, representation, budget and transparency findings for a council, newspaper or regulator. In development.

Your transcript and a copy of your saved perspective live in your browser, not on our servers. Pages you have visited can load from cache on a flaky connection. The interview itself needs a connection, because the mediator runs on the server.

High contrast mode. Large text mode. Reduced motion mode. Keyboard navigation. Screen reader labels. Skip-to-content links. Voice input. Because a civic tool that excludes people isn't civic.

Add Civic Filament to your phone's home screen and open it full-screen like an app. No app store required. Native iOS and Android apps are on the roadmap.

Published syntheses and their verification data are available as JSON, CSV and Markdown, plus a print-ready council report, so others can check them, file them or build on them.

Generate short and long narratives about a community's findings in different tones. In development.

After each synthesis, identify what the system doesn't know yet and steer future interviews to fill those gaps. In development.

Participation doesn't require an account. If you want one, sign up with email and password to link your perspectives to it. Forgot your password? A reset link goes to your inbox. Want to leave? One button permanently deletes your account and everything linked to it — no support ticket required.

A daily job pulls new Executive Orders from the Federal Register so a community can deliberate on a fresh policy the same week it lands. Proposed rules go to a human-review queue rather than auto-publishing. Beta: the daily job is scheduled; we are still confirming it runs reliably in production.

Verification and password-reset emails sent from a civicfilament.org address instead of a generic sender.