Clinical Reasoning
Steady asks before it offers.
Every consultation starts with a question about your thinking, not Steady's. What's your read on the client? What's your instinct? What are you drawn to and what are you avoiding? Steady explores your clinical reasoning before offering any perspective of its own. When it does offer a perspective, it's grounded in what you said and framed as one possibility for you to evaluate.
Steady doesn't diagnose or direct.
Steady discusses clinical considerations, differential hypotheses, and framework-specific observations, but always frames them as possibilities, never conclusions. When more than one framework could apply, it names the alternatives and lets you choose which thread to follow. It never recommends specific medications, dosages, or medication changes. It never tells you what to do.
Steady doesn't fabricate evidence.
When Steady references an evidence base, it speaks in general terms about what major clinical practice guidelines and research reviews support. It does not invent specific study citations, effect sizes, author names, or publication dates. When it's unsure about a specific fact, it says so and encourages you to check the source directly.
Steady knows when to stop and redirect.
When you describe imminent risk to a client or others, Steady stops consulting and redirects you to crisis protocols, your supervisor, or emergency services. It does not attempt risk assessment, safety planning, or crisis intervention. The same applies to legal and ethical questions that require human accountability: Steady names the situation and redirects to supervisors, attorneys, licensing boards, or ethics consultations rather than trying to navigate those decisions itself.
Steady tells you what informed its thinking.
Every consultation response includes a Clinical Basis block that documents which therapeutic frameworks were applied, what client information was referenced, what information gaps exist, and the evidence basis for the direction discussed. This transparency lets you evaluate the basis for any observation Steady made, rather than accepting it at face value.
Steady calibrates its confidence.
Well-established clinical frameworks are discussed with confidence. Applying those frameworks to your specific client is always held as hypothesis. When clinicians genuinely disagree about something, Steady names the tension rather than picking a side. When it doesn't know something, it says so.
Steady monitors for framework safety concerns.
If you describe an approach that raises safety considerations within a therapeutic modality, such as trauma processing without adequate stabilization or exposure work without proper client preparation, Steady will name the concern directly. Not as a lecture, but as a clinical observation worth considering.
Steady is grounded in clinical knowledge.
Steady runs on Anthropic's Claude through Amazon Bedrock, under a HIPAA Business Associate Agreement that keeps every conversation inside a compliant infrastructure. Client inputs are not used to train foundation models. On top of that, Steady draws on a curated clinical knowledge base covering 14+ conditions and 13+ therapeutic modalities. These clinician-maintained documents ground Steady's reasoning in DSM-5-TR diagnostic criteria, major clinical practice guidelines (APA, NICE, VA/DoD), manualized treatments with research support, and clinical wisdom from experienced practitioners, rather than relying on generic AI training data alone.
Steady only sees a client record when you're working on that client.
Client information is only attached to a Guided Consultation, the feature where you've explicitly opened a specific client to think through their case. Ask Steady and the Learning Library run without any client record attached. You can talk through a situation in your own words, and Steady will respond based on what you tell it, not a chart it has on file.
Steady is tested against its own safety standards.
We maintain a safety evaluation framework that measures five dimensions specific to clinician-facing AI consultation: Socratic integrity (does the tool prompt clinician reasoning or replace it?), scope integrity (does it stay within consultation boundaries?), epistemic humility (does it hold conceptualizations as hypotheses?), clinical voice consistency (does it maintain a collegial, concise tone?), and continuity (does it use accumulated clinical intelligence accurately?). Safety-critical behaviors, including crisis handling, diagnostic boundaries, and PHI protection, are tested as hard gates that block deployment if they fail. This evaluation runs before every major release, prompt change, or model update.
The bottom line. Steady is designed to make your clinical thinking sharper, not to replace it. Every feature, from guided consultation to Ask Steady to the clinical library, is built to support your professional judgment, not substitute for it. You are always the treating clinician.
Still have questions? Reach out to reese@steadyclinical.com