Explore candidate matches against the 79-entry research taxonomy.
79 pattern entries loaded locally. Bounded conversation data is saved in this browser and restored on reload. In cloud mode, your chat messages and API key are sent to the selected model provider. A local (WebGPU) model sends no conversation to an AI provider. Clear the saved conversation with New Session.
⚠ Experimental: This clinic is a research preview. Its diagnostic output is exploratory and has not been clinically validated. Do not rely on it for consequential decisions about AI systems without independent verification.
Session settings
This is a research preview. The 79 pattern entries are loaded locally in your browser.
If you choose a cloud model, the complete provider request, including your chat messages, system instructions, tool definitions, tool calls and results, and API key, goes directly to Anthropic's servers. The key is held only in this page's memory and cleared on reload or tab close. Bounded structured conversation data is saved in localStorage for reload recovery. Clear it with New Session, or by clearing this site's data. The semantic-search runtime is self-hosted; its model files download from Hugging Face only when semantic search is first used. The model host sees your IP address but receives no Clinic conversation text.
If you choose a local (WebGPU) model, your chat messages stay on your device and are not sent to an inference provider. The executable WebLLM and semantic-search libraries are self-hosted. Model weights and compiled model files download from the registered Hugging Face and GitHub hosts, which receive your IP address but no Clinic conversation text. Downloaded assets may remain in browser-managed Cache Storage or IndexedDB until you use Delete cached model or clear this site's data. No API key is needed.
Research preview, not for clinical use. All 79 Pattern-guidance entries and evidence records await independent expert review. Axis 10 (entries 10.1–10.15), including Hybrid Pathologies 10.4–10.15, was ratified into the canonical taxonomy in v2.2 (June 2026). That taxonomy decision does not approve the LLM-drafted probes, rubrics, or interventions. Diagnostic output from this tool does not constitute expert clinical opinion.
Read the responsible-use boundaries before testing or acting on a result.
An API key is required to start a cloud session. Paste your key above, or choose a local model.
Get one at platform.claude.com. The key stays in this tab's memory and is cleared on close.
Cloud models require an API key. The local model (Qwen 3 8B) needs none; the first run downloads its weights (~5 GB). Running it needs more memory than the download alone: the 32k context window adds a multi-gigabyte KV cache, so it needs roughly 8–10 GB of GPU or unified memory in total. Its memory floor is shown when you select it.
Start a new session?
This will clear the current conversation and provenance trail.