AI Transparency & Responsible AI Statement

1. Responsible AI principles

• Useful, not autonomous. AI features are designed to augment construction professionals, not to replace their judgment. A human reviews and decides.
• Consent before surprise. New AI features are clearly labeled; voice, multimodal, and data-sharing behaviors require opt-in.
• Don't train on your business. We do not contribute Customer Data to foundation models that we do not own. See Section 5.
• Traceability. Significant AI actions are logged in the Workspace audit trail so a human can reconstruct what happened.
• Graceful degradation. If an AI provider is unavailable, the product continues to work without it — we do not hide critical non-AI workflows behind AI.

2. Where AI is used

Today, Chantro uses AI in features such as:

• Conversational assistant. A chat interface that can answer questions, draft short artifacts, and invoke a limited set of workspace tools — for example, creating calendar events, tasks, or project items — that you confirm.
• Voice mode. A speech-to-text and text-to-speech interface that turns spoken instructions into assistant actions.
• Summarization. On-demand summaries for project updates, task discussions, and related workspace content.
• Search grounding. The assistant can look up information from your workspace or, when permitted, from the web to improve the quality of its answers.

We may add AI features over time — for example, proposal drafting, document intelligence, or anomaly detection. When we add them, we will update this statement and note what data they send to which provider.

3. Model providers

We route each request to the provider best suited to the task, balancing quality, latency, and cost. Our current providers, and the commercial terms they operate under, are listed in our Subprocessor List. For each provider, we configure the API to prohibit training on your prompts and outputs where that provider's tier supports such controls. Providers may retain content briefly for abuse monitoring as described in Section 4. We are moving providers onto paid and, where available, enterprise or zero-retention tiers as part of our production hardening.

4. Data flow & retention

• What is sent: the prompt you write or the feature parameter you trigger, plus the specific workspace context needed to answer it (for example, the project or document you are asking about). We do not send the entire workspace.
• Where it is processed: at the relevant model provider (see Subprocessor List). Providers' endpoints are typically located in the United States.
• How it is retained: we store the prompt and the AI Output in your Workspace (so you can read, search, and share it, and so admins can audit use). We do not keep a separate copy with providers unless their abuse-monitoring policy requires a short window (commonly ≤ 30 days) and we are unable to disable it under enterprise terms.
• Voice mode: audio streams are transcribed by our speech-to-text provider and then discarded at the provider; the transcript is stored in your Workspace.

5. Training & fine-tuning

• We do not contribute Customer Data to train foundation models owned by OpenAI, Anthropic, Google, or any other third-party model provider, and we configure provider APIs to honor that intent where their tier supports it.
• We may train or fine-tune models that we own (or run on private infrastructure) using Customer Data only where permitted by contract, aggregated or per-Workspace. For a per-Workspace fine-tune, the resulting model benefits only that Workspace.
• We use aggregated, de-identified statistics about feature usage to improve quality (for example, which prompts needed retries, which outputs received thumbs-down).

6. Customer controls

• Disable AI features. Individual Users can disable the conversational assistant, voice mode, and related AI features from their personal settings. Workspace-level administrative kill switches are on our near-term product roadmap.
• Opt out of feedback sharing. By default, thumbs-up/thumbs-down feedback on AI Output is aggregated and used to improve the product. You can decline to leave feedback at any time.
• Credits & caps. Users can see AI credit consumption in their billing and usage views and set personal daily caps.
• Data residency. Today, AI requests are processed at the model provider's U.S. endpoints. Region-specific routing is not yet offered; when we add it, it will be configurable at the Workspace level.

7. Human oversight

AI Output is presented as a draft, suggestion, or decision-support signal. The product does not execute financial, contractual, or safety-critical actions without a human confirmation step. When AI suggestions are accepted, the identity of the accepting User and the underlying AI context are recorded in the audit trail.

8. Accuracy & hallucinations

AI systems can produce incorrect, outdated, or fabricated information (commonly called “hallucinations”). We work to reduce hallucinations by grounding the assistant in your Workspace content, surfacing citations where we can, and running evaluation sets on our prompts. Even so, you must verify any AI Output before using it for a bid, contract, change order, purchase, schedule commitment, or safety decision. We are not liable for decisions made on the basis of unverified AI Output — see the Terms of Service §6 and §17.

9. Prohibited content & misuse

AI features are subject to the Acceptable Use Policy. Attempts to use AI features to generate disallowed content, impersonate others, evade safety filters, or extract model internals may result in suspension or termination.

10. Use by minors

Chantro is not intended for use by children under 16. Workspace Owners are responsible for ensuring Users meet the eligibility requirements in the Terms of Service §3.1.

11. Feedback and appeals

We welcome reports of problematic AI behavior — incorrect outputs, unsafe outputs, or disproportionate impacts on a group. Email support@chantro.com with the feature, the prompt, and the output, and we will investigate. Where an automated system has suspended or restricted a Workspace based on AI-detected behavior, Owners may request human review at trust@chantro.com.

12. Regulatory commitments

We track emerging AI regulation — including the EU AI Act, U.S. state AI laws, and sector-specific rules — and update our features, documentation, and disclosures to comply. Where a Customer's regulated use case imposes extra obligations (for example, impact assessments), we will provide reasonable cooperation.

13. Updates to this statement

As we evolve our AI features and add providers, we update this statement. The “Last updated” date at the top of this page reflects the most recent review.

14. Contact

Questions about AI features, model providers, or this statement: privacy@chantro.com.