Talk to the City

Free (Open Source) Tier Ratings

Last Reviewed: 2026-03-23

Primary use case. Processes hundreds to thousands of responses while preserving nuance. Used by Taiwan government, unions, advocacy orgs.

Excellent for large-scale qualitative input. Young Women's Alliance Australia used T3C to analyze 85 interviews, leading to legislation.

Healthcare workers union used T3C to poll members for contract negotiations. Captures depth beyond simple polls.

Can synthesize diverse stakeholder perspectives for strategic planning. Interactive reports help leadership see themes and drill into specifics.

Core strength. Taiwan Ministry of Digital Affairs hosts own instance for national consultations on AI policy, elections, social issues.

CMI used T3C in Yemen for youth voices on peacebuilding with 94% response rate across 18 governorates.

Can help identify shared concerns across diverse groups. Works well when you have substantial input data to analyze.

Recursive Public project with Chatham House engaged 1,000+ participants on AI governance, funded by OpenAI Democratic Inputs grant.

Can analyze qualitative program feedback at scale. Less suited for quantitative outcome measurement.

Never (OpenAI API)

Uses OpenAI API (default: gpt-4o-mini); additional LLM providers planned. OpenAI explicitly does not use API-submitted data to train or improve models (policy in effect since March 1, 2023; confirmed as of 2026). No training use by T3C itself. Data is also sent to Google Cloud Platform, Posthog (analytics), and Weights & Biases (model monitoring) for processing — none use it for training.

Not needed

OpenAI API data is never used for training by default — no opt-out required. For self-hosted deployments, you bring your own API keys and control all data flow entirely.

Never (routine)

No routine human review of your data. All processing is automated via LLMs. OpenAI retains abuse monitoring logs for up to 30 days for policy enforcement purposes, but this is standard API use — no human reviews report content. T3C itself does not review submitted data.

User controlled (hosted) / Full control (self-hosted)

Hosted version (talktothe.city): Reports are stored on the platform until deleted by the user. Currently in alpha — reports are public to anyone with the exact URL (password protection is planned but not yet implemented as of March 2026). OpenAI abuse monitoring logs retained up to 30 days per OpenAI policy. Self-hosted: You control all data and retention.

Open source auditable; no formal certifications

No SOC 2, ISO 27001, HIPAA, or FedRAMP certifications. Open-source code (Apache-2.0 license) is fully auditable at github.com/AIObjectives/tttc-light-js. Privacy policy last updated September 18, 2025. Data processors include OpenAI, Google Cloud Platform, Posthog, and Weights & Biases. The vendor states data is not shared, sold, or distributed beyond these processors. GDPR/CCPA rights acknowledged (EU/UK/California users). Age restriction: not for users under 13; 13–16 may require parental consent.

Genuinely Free and Open Source

No freemium tricks. No paywalls. Built by a 501(c)(3) nonprofit (AI Objectives Institute) with funding from Google.org, Future of Life Institute, and others. All code on GitHub (Apache-2.0 license) at github.com/AIObjectives/tttc-light-js.

Hallucination Mitigation by Design

Every AI-generated claim links directly to verbatim participant quotes. You can verify any summary against the original statements. This is the core design principle, not an afterthought.

Proven at Government Scale

Taiwan Ministry of Digital Affairs hosts its own T3C instance for national consultations. Used for AI policy, same-sex marriage debates, election platform analysis. This isn't a startup experiment.

Specialized Tool, Not General Purpose

T3C does one thing well: analyzing large-scale qualitative input. It's not for daily tasks like writing emails or creating graphics. Best with 100+ responses to analyze.

Full Audit Trail

Every processing decision is logged: filtering, deduplication, extraction results, timestamps, model versions. If something looks wrong, you can trace exactly what happened.

WhatsApp Integration

Collect responses via WhatsApp conversations with AI-guided structured dialogue. Dramatically lowers barriers for participation, especially in low-resource contexts (used in Yemen peacebuilding project).

2026-03

Talk to the City 2026 promotional video published by AI Objectives Institute, signaling continued active development and outreach.

2025-09

Privacy and Security Policy updated (effective September 18, 2025). Now explicitly names data processors: OpenAI, Google Cloud Platform, Posthog, and Weights & Biases.

2025-02

Good Ancestor Foundation milestone recognition: major codebase refactor completed; active development migrated to tttc-light-js GitHub repository (old talk-to-the-city-reports repo being archived).

2025

New hosted platform at talktothe.city with streamlined report creation; default model is gpt-4o-mini; additional LLM providers announced as planned.

2024

Australia Young Women's Alliance study led to legislation passed by New South Wales Legislative Council.

2024

Tokyo gubernatorial election deployment with 1,000 residents, helping independent candidate achieve record vote share.

2024

Yemen deployment by CMI for youth voices on peacebuilding across 18 governorates.

For organizations with 5+ staff

Excellent choice for organizations that regularly conduct large-scale stakeholder consultations, public input processes, or membership surveys. The open-source model with no licensing costs means no procurement friction. Dedicated staff can self-host for full data control. Plan for setup time and staff training — this is a specialized analysis tool, not a turnkey survey platform.

For organizations with fewer than 5 staff

Highly accessible via the hosted version at talktothe.city — no installation required. Free with no account limitations. Best when you have a specific consultation project with 100+ responses to analyze. Not a daily-use AI assistant. Note that in the current alpha, hosted reports are publicly accessible by URL (no password protection yet), so avoid uploading sensitive participant data to the hosted version until this is resolved.

For health data or vulnerable populations

Use with caution. No HIPAA BAA is available. No SOC 2 or formal security certifications. The hosted version currently makes reports public by URL (password protection planned but not yet implemented). For projects involving health information, domestic violence survivors, or other sensitive populations, strongly consider self-hosting for full data control, or anonymize all participant data before upload. Do not submit protected health information (PHI) to the hosted platform.

Best for

Advocacy organizations, foundations, membership associations, and government agencies that need to gather and analyze large-scale stakeholder input while preserving individual voices. Ideal when you have 100+ responses to analyze and need transparent, auditable AI analysis.

Watch out

Specialized tool — not for daily AI tasks. Best with 100+ responses. No SOC 2, HIPAA BAA, or FedRAMP certification. Uses OpenAI API by default (gpt-4o-mini); additional providers planned. Hosted reports are currently public by URL in this alpha phase (password protection coming). Additional data processors include Google Cloud Platform, Posthog, and Weights & Biases. Learning curve required. Community support only — no enterprise SLA.

"Back in 2014, it was impossible to interview a mini public of people and aggregate their ideas while preserving the full nuance. But now, with Talk to the City's help, that cost has been reduced to essentially zero. It's broad-listening, and it can change the nature of this recursive public."

"Talk to the City's analysis was crucial in identifying gender-specific challenges... Having the data organized under specific themes, then being able to go through the interviewees' actual statements, was powerful."