Support with clear boundaries.

What Forge is designed to do

Forge helps users slow down, name what is happening, consider practical options, and reflect without judgment. It avoids fake positivity, manipulative engagement, toxic masculinity, invented personal stories, diagnosis, and claims of human experience.

Layered safety checks

• Deterministic input rules identify known crisis, self-harm, violence, abuse, and other high-risk language.
• An AI safety classifier can escalate risk when deterministic checks do not already block normal coaching.
• High-risk or crisis classifications stop the ordinary coaching response and switch to direct safety guidance.
• Generated responses are screened against output-safety rules before delivery.
• Published prompts, guidance, and custom guardrails are controlled through restricted administrative workflows.

Memory and privacy

Memories can help Forge provide more relevant support, and you can manage or delete them through your data controls. See the Privacy Policy for more about how your information is used.

Crisis boundary

ForgeMind is not monitored emergency care and cannot contact emergency services for you. If there is immediate danger, contact local emergency services now. In the United States, call or text 988. Where possible, move to a safer place and contact a trusted person nearby.

Known limitations

No keyword rule, classifier, or language model can understand every context or guarantee a safe or accurate response. Sarcasm, coded language, missing context, provider outages, and model errors can produce false alarms or missed risk. Guardrails reduce risk; they do not remove it.

Human in the loop

Users decide whether to follow a suggestion and should verify high-stakes information with a qualified professional. Authorized operators can review safety behavior, publish prompt changes, and update guardrails. Product analytics and safety records are governed by the Privacy Policy and Data Safety summary.

Report a concern

If Forge produces an unsafe, harmful, or misleading response, stop relying on it for that situation and contact us. Do not include sensitive details unless they are necessary to investigate the issue.