Bria Clinical AI: Ambient scribe, coding assistance and clinical risk prediction

Ambient AI clinical scribe (Nuance DAX / Abridge / Suki / Augmedix parity) — Claude primary LLM

Add it to your plan Start free trial

+$38.40/mo at 20 practitioners. Free instant access, no credit card required.

Works with

Electronic Health Record

What's inside

Everything in Bria Clinical AI

42 items in 8 areas, every one shipping with the app today.

Encounter Recording & Capture
- Encounter recording recordsEach clinician-patient encounter is captured as a recording record with patient, clinician, encounter reference, audio URL and MIME type.
- Auto-numbered recordingsEvery recording gets a unique sequential reference (AI/ENC/...) from a dedicated numbering sequence at creation.
- 8-language + auto-detect supportRecordings can be tagged English, Swahili, French, Arabic, Zulu, Yoruba, Hausa, Amharic or auto-detect for transcription.
- Automatic duration trackingEncounter duration in seconds is computed and stored from start and end timestamps.
- Patient consent flagA consent-recorded checkbox documents that the patient agreed to being recorded before AI processing.
- Retention date controlA retain-until date on each recording supports data-retention policy enforcement.
AI Note Drafting (Scribe)
- One-click SOAP drafting via ClaudeA Draft SOAP action calls the Anthropic Messages API (claude-sonnet-4-5 default) with the transcript and an active SOAP template to generate the note.
- Configurable model parametersTemperature, max tokens and system/user prompts come from the selected template and provider, sent verbatim in the API request.
- Graceful stub fallbackIf no API key or active SOAP template is configured, a deterministic offline SOAP stub is generated so the workflow never blocks.
- Error-safe API handlingHTTP errors and exceptions are caught, logged on the task with the message, and marked failed without ever raising to the user.
- Real-time cost accountingInput/output token usage from each API response is converted to USD cost using built-in per-model Anthropic pricing tables.
- Latency measurementEvery LLM call records wall-clock latency in seconds on its task for performance monitoring.
Clinical Workflow & Sign-off
- 8-stage encounter state machineRecordings move through Recording, Uploaded, Transcribing, Drafting, Draft Ready, Clinician Reviewing, Signed Off and Discarded with a statusbar.
- Stage-aware workflow buttonsMark Uploaded, Start Transcription, Mark Draft Ready, Sign Off and Discard buttons appear only in valid states.
- Separate transcription and LLM statusesIndependent pending/in-progress/completed/failed statuses track speech-to-text and note-generation pipelines.
- Clinician sign-off gateAn explicit highlighted Sign Off action closes the human-in-the-loop review before the note is final.
- Human edit audit logEach edit is logged with editor, timestamp, field changed, original vs new text, edit distance and accepted-AI flag for HITL benchmarking.
AI Task Ledger & Monitoring
- Per-call scribe task recordsEvery LLM or transcription job creates a task record linking recording, provider and template with full prompt and response text.
- Task lifecycle statesTasks track queued, running, succeeded, failed, canceled and retrying statuses with start/finish timestamps and retry counts.
- Prompt/response/error inspectionTask form exposes the exact JSON prompt sent, the raw response and any error message in notebook tabs.
- Automatic failed-task retry cronA 5-minute scheduled job requeues failed tasks (up to 3 retries, 50 per pass) into retrying state.
- Task cost and token analyticsPivot and graph views break down cost, prompt tokens and completion tokens by provider and status.
Usage Quotas & Cost Control
- Per-provider usage quotasQuota records track requests made vs limit, tokens in/out and total USD cost per provider per period.
- Daily usage aggregation cronA nightly job rolls up yesterday's tasks per provider into quota rows (requests, tokens, cost), creating or updating as needed.
- Quota threshold alerting cronAn hourly job posts a chatter alert once usage crosses the configurable threshold percentage (default 80%) and marks it sent.
- Cost pivot analysisA pivot view slices total USD cost by provider and period for spend review.
AI Provider Registry
- Multi-vendor provider catalog11 provider types: Anthropic Claude, OpenAI GPT, Azure OpenAI, Google Gemini, AWS Bedrock, Ollama, llama.cpp, Whisper local/API, Deepgram, Azure Speech.
- Capability routingEach provider declares a capability (transcription, summarization, coding/billing, translation, dictation, multimodal) used to pick the right engine per job.
- PHI-safe (BAA/DPA) flagProviders can be marked PHI-safe when a Business Associate or Data Processing Agreement exists, with a search filter to find them.
- Rate limits and defaults per providerRequests-per-minute, temperature, max tokens, model name, endpoint and region are configurable per provider with unique-code enforcement.
- Key-never-stored designProvider records carry only an API-key-configured boolean; the actual secret lives in system parameters, never on the record.
- 8 pre-seeded providersShips Claude Sonnet 4.5, Claude Opus 4.1, GPT-4o, Azure OpenAI (HIPAA), Gemini 2.5 Pro, Bedrock Claude, Whisper API and Deepgram Nova-3 Medical.
Prompt Template Library
- Prompt template engineTemplates pair a system prompt with a user prompt that takes the transcript and the patient's details, plus per-template temperature and max tokens.
- 8 note output formatsTemplates target SOAP note, HPI only, discharge summary, operative note, referral letter, admission note, progress note or structured JSON.
- 15 medical specialtiesTemplates scope to General, Paeds, OB/GYN, Surgery, ICU, Emergency, Cardio, Onco, Derm, Psych, Ortho, Uro, MCH, HIV/ART and TB.
- 10 ready-made clinical templatesShips SOAP, HPI, Discharge, Operative Note, Referral, Admission, Progress, Paeds SOAP, OB/GYN SOAP and strict-JSON extractor with anti-hallucination prompts.
- Per-template redaction requirementA requires-PHI-redaction flag marks templates whose input must be de-identified first.
PHI Redaction & Privacy
- Rule-based PHI redaction engineRedaction rules combine regex, named-entity or list-lookup matching with custom replacement text and drag-to-order sequencing.
- 9 PHI classes coveredRules target patient name, DOB, MRN, phone, email, address, SSN, national ID or all PHI.
- Severity tiers and language scopingEach rule is Required, Recommended or Optional and can be limited to any of 8 languages or auto.
- 12 pre-seeded redaction rulesShips US SSN, Kenya national ID, Nigeria NIN, Kenyan and E.164 phone, email, two DOB formats, MRN, street address, credit card and person-name NER rules.
- PHI-redacted flag on recordingsRecordings carry a redacted boolean so downstream users can see whether de-identification was applied.

Works with

What Bria Clinical AI works with

Bria Clinical AI shares one database with the rest of the Bridge HMS Suite and BridgeERP: these are the apps it already talks to, with nothing to buy in between.