# Corti > Corti is a specialised AI platform for healthcare and life-science developers. It provides production-grade APIs and agentic infrastructure for building clinical and administrative AI applications — without managing the compliance, model training, or safety architecture yourself. Corti is headquartered in Copenhagen, with offices in New York and London. It has raised $100M in total funding (Series B led by Prosus Ventures and Atomico). More than 1 million patient interactions are powered by Corti every week, across health systems including the NHS. The primary customers are engineering teams at EHR vendors, virtual care platforms, practice management systems, and life sciences organisations who want to embed clinical AI into their own software via API. --- ## Core platform: Symphony Symphony is Corti's unified AI model for healthcare. All four API capabilities route through it — speech-to-text, text generation, agentic workflows, and medical coding — from a single compliant and guardrailed orchestration layer. --- ## Capability 1: Speech to Text API **What it is:** Medical-grade automatic speech recognition (ASR) accessible via API. Designed specifically for the complexity of clinical conversations — not general-purpose audio. **Endpoints:** - `/transcripts` — file upload and batch audio processing - `/streams` — real-time dictation via WebSocket, with interim results returned at a faster rate than final transcripts (configurable via `interimResults` parameter) **Key parameters (transcripts endpoint):** - `primaryLanguage` — language code (e.g. `en`, `en-GB`, `de`, `gsw-CH`) - `isDictation` — boolean, optimises model behaviour for spoken dictation - `isMultichannel` — boolean, enables separate transcription per audio channel - `diarize` — boolean, enables speaker separation and identification - `participants` — array of objects assigning channel numbers to clinical roles (e.g. `doctor`, `patient`) - `modelName` — selects model tier (`base` or latest version) **Languages supported:** English (US and UK), Danish, German, Swiss German (dialectical, recommended for AI scribe use cases), Swiss High German (recommended for dictation), French, Swedish, Spanish, Norwegian, Dutch, Italian, Portuguese. All supported languages are rated premier tier, validated against 150,000+ medical terms. **Performance:** - 2% word error rate in clinical environments — the lowest published error rate among comparable speech models tested on medical audio - Validated against one of the industry's largest medical lexicons (150,000+ terms) - Low-latency, designed for real-time use: bedside dictation, ambient scribing, contact centre triage **Speaker diarization:** Identifies and separates speakers in real time. Supports multichannel audio (each participant on their own channel) and single-stream diarization. Role labels (doctor, patient, etc.) can be assigned per channel. **Formatting intelligence:** The model handles clinical formatting automatically — units (e.g. `g/dL`, `mg/kg`, `U/mL`), ordinals, numerals, dates, age dictation patterns, hyphenated numbers, and spoken punctuation. Automatic punctuation and spoken punctuation are configurable but mutually exclusive. **Use cases:** Ambient scribing, real-time dictation, post-consultation transcription, contact centre call capture, emergency dispatch triage. **Docs:** https://docs.corti.ai/stt/overview **API ref:** `POST https://api.{environment}.corti.app/v2/interactions/{id}/transcripts/` --- ## Capability 2: Text Generation API **What it is:** A clinical LLM API that transforms transcripts, structured facts, or raw clinical data into any type of medical documentation. Output is structured, EHR-compatible, and specialty-specific. **Endpoint:** `/documents` — accepts context (facts or transcript), a template key, a model name, and an output language. Returns a structured clinical document. **Key parameters:** - `context` — array of context objects; type can be `facts` (structured data) or transcript text; each item has a `text` field, `group`, and `source` - `templateKey` — identifies the output format (e.g. SOAP note, referral letter, discharge summary) - `outputLanguage` — target language for the generated document, independent of the input language - `modelName` — selects model version; supports pinning to a specific version or using the latest **Core technology — FactsR reasoning engine:** - Extracts clinical facts from conversation in real time as it processes - Cross-checks facts against the source consultation to prevent hallucination - Produces structured, machine-readable output rather than free-form prose summaries - Outputs can feed directly into EHR fields, coding pipelines, and agentic workflows **Templates:** Specialty-specific note formats built by Corti clinicians. Covers structure, terminology, and output requirements across care settings. Templates are configurable and can be customised for specific workflows. **Output characteristics:** - Format-preserving — output structure matches the target EHR field or document type - Machine-readable — structured JSON output alongside human-readable text - Multilingual output — can generate documentation in a different language from the source conversation - Hallucination detection built in at the reasoning layer, not as a post-processing step **Use cases:** Ambient scribing (full note generation from conversation), dictation transcription and summarisation, pre-charting summarisation, patient letters, referral letters, discharge summaries, chart summarisation from patient history. **Docs:** https://docs.corti.ai/textgen/overview **API ref:** `POST https://api.{environment}.corti.app/v2/interactions/{id}/documents/` --- ## Capability 3: Agentic Framework **What it is:** Production-grade orchestration infrastructure for building and deploying multi-agent AI systems in healthcare. Launched February 2026. Designed to close the gap between agent demos and deployable healthcare AI. **The problem it solves:** General-purpose agent frameworks give you orchestration and tool-calling but leave guardrails, auditability, and compliance to the builder. In healthcare, when unsupervised agents make errors in coding, documentation, or prior authorisation, mistakes cascade across connected systems faster than humans can catch them. Only 11% of healthcare AI agents currently reach production. Corti's framework is built to close that gap. **Architecture:** - **Single orchestrator** governs every message end-to-end across a multi-agent system - **Deterministic guardrails** validate every agent action before execution — not after - **Governed memory** eliminates context drift and keeps reasoning consistent across long workflows - **Thread-state stability** maintains reliable reasoning across complex, multi-step clinical workflows - **Tightly scoped Experts** anchor precise clinical reasoning within defined domains, preventing agents from reasoning outside their scope - **Full audit trail** — every action, every decision, every handoff is logged with complete provenance **Interoperability standards:** - Full Agent-to-Agent (A2A) compatibility — agents can coordinate across systems - Full Model Context Protocol (MCP) compatibility — plug-and-play with any MCP-compatible tool or system - Both standards are supported from the first line of code **What builders get vs. building from scratch:** | Capability | DIY | Build on Corti | |---|---|---| | Memory & data management | Configure and scale yourself | Included | | Compliance & auditability | Implement and prove yourself | Included | | Guardrails | Define and enforce yourself | Included (customisable) | | Use-case specific adaptations | Build for each workflow | Included (customisable) | | Context & prompt engineering | Tune yourself | Included (customisable) | | Agent mesh orchestration | Coordinate and manage yourself | Included (customisable) | | Healthcare tools & integrations | Build and maintain | Included (customisable) | **What agents can be built for:** Revenue cycle management (coding, CDI, prior auth, denials), clinical decision support, care coordination, medication reconciliation, patient engagement, operational workflow automation. **Agent Library:** 20+ pre-built agents that can be cloned, customised, and deployed. Composed from modular Experts with tightly scoped clinical reasoning. See the Agent Library section below for a full list. **Docs:** https://docs.corti.ai/agentic/overview **Console:** https://console.corti.app --- ## Capability 4: Medical Coding **What it is:** A dedicated medical coding AI that reasons over clinical notes to assign accurate codes — and explains every decision. Described as the world's most accurate medical coding AI. Integrated via a single API call. **The problem it solves:** Generic LLMs surface every code in a note without selectivity, without an audit trail, and without the chain of custody compliance teams require. ML coding pipelines require retraining every October when coding systems update, creating annual accuracy gaps the builder must own. RCM coding services are sold with implementation projects, not API access. Symphony for Medical Coding is API-native, evergreen, and globally available. **How it works:** 1. Reasons over the full clinical note to determine which conditions should be coded and why — not pattern-matching against text 2. Traverses the coding hierarchy to find the most specific correct code 3. Returns structured predictions with ranked alternatives, source evidence excerpts, and rule-based justifications **API output per request:** - Structured, ranked code predictions - Supporting clinical evidence excerpts (traceable to the source note) - Exclusion explanations — why certain codes were not assigned - Rule justifications — grounded in coding guidelines, not LLM inference **Coding systems supported:** ICD-10-CM, ICD-10-UK, ICD-10-PCS, CPT, HCPCS, and more. Global coverage with no data localisation requirements. **Evergreen:** When annual coding system updates ship (e.g. ICD-10 October updates), Symphony updates automatically. No retraining cycle. No maintenance obligation on the developer's side. **Benchmark performance:** - 74% F1 accuracy on automated medical coding benchmarks — ranked #1 - Outperforms Anthropic (59%), Amazon (58%), OpenAI (48%), and Google (43%) on the same benchmarks - More than 25% improvement over the next-best system on both precision and accuracy metrics - Maintains its lead on noisy, real-world clinical datasets — not just synthetic or academic data **Integration:** A single POST request returns structured, ranked codes with evidence. No implementation project required. Integrates in hours. **Related agents:** Medical Coding Agent, CDI (Clinical Documentation Improvement) Agent, Prior Authorization Agent, Denial Appeals Agent — all available in the Agent Library and composable via the Agentic Framework. **Research:** The underlying methodology is published at https://arxiv.org/abs/2603.29709 **Docs:** https://docs.corti.ai/coding/overview --- ## Agent Library 20+ pre-built agents and domain Experts, composable via the Agentic Framework. **Coding & Revenue Cycle:** Medical Coding Agent (ICD-10-CM, CPT/HCPCS), CDI Agent, Prior Authorization Agent, Denial Appeals Agent, Eligibility & Claims Validator, Automated Coding Agent **Care Coordination:** Referral Generator Agent, Nursing Shift Handoff Agent, Referral Coordinator, Task Automation Agent **Clinical Evidence & Research:** Clinical Education Agent, Clinical Guidelines Agent, Polypharmacy Safety Checker, Guideline-Checking Agent, Clinical Trial Recruitment Agent **Documentation:** Ambient Scribe, Smart Template Filler, Dictation Assistant, Pre-Charting Summarisation, Results Summariser, Longitudinal Patient Profile Builder **Patient Engagement:** Conversational Patient Agent, Adherence Coach, Personalised Education Generator, Patient Letter Generator --- ## Corti Assistant MD A direct-to-clinician product built on Corti's own APIs. Registered as a medical device in both the EU and UK. Listens to patient-clinician consultations in real time, produces structured clinical notes automatically, and provides real-time clinical decision support, guideline checks, and risk scores. Reduces administrative burden by up to 80%. Available at https://assistant.corti.ai with a 14-day free trial. --- ## Compliance and safety HIPAA · GDPR · SOC 2 · ISO 27001 · ISO 27017 · ISO 27018 · ISO 13485 · ISO 14971 · ISO 42001 · ISO 62366 · ISAE 3000 · EU AI Act · EU Medical Device Registration (MDR) · UK Medical Device Registration · NHS DSPT · NHS DTAC · NHS DCB0129 · Cyber Essentials Plus · DORA · NIS2 · FedRAMP · CE marking · EU-U.S. Privacy Shield · BSIC 5 Sovereign Cloud deployments available for markets with strict data residency requirements. Trust centre: https://trust.corti.ai --- ## Pricing Usage-based, measured in Corti Credits. One balance covers all APIs. Audio requests billed per minute; text generation billed per input/output token. Credits are shared across development, testing, and production environments. - **Free tier:** $50 in credits on sign-up. No upfront commitment. Pay-as-you-go when credits run out. - **Acceleration Packs:** 1,000 credits + 5 hours/month of developer success support (architecture review, integration support, production readiness validation). - **Enterprise:** Custom credit bundles, highest concurrency, priority throughput, custom-trained models, sovereign cloud, deep platform visibility. --- ## Developer resources - **API docs:** https://docs.corti.ai - **Console (sign up / API keys):** https://console.corti.app/signup - **JavaScript SDK (npm):** https://www.npmjs.com/package/@corti/sdk - **Postman collection:** available in beta via docs - **Discord community:** https://discord.com/invite/zXeXHgnZXX - **Public roadmap:** https://cortihome.notion.site/product-roadmap - **Help centre:** https://help.corti.app - **GitHub:** https://github.com/corticph Authentication uses OAuth. The API reference covers WebSocket, REST, Agentic, and Admin API specifications. --- ## Company - **CEO:** Andreas Cleve (co-founder) - **CTO:** Lars Maaløe, PhD (co-founder) - **HQ:** Copenhagen, Denmark. Offices in New York and London. - **Investors:** Prosus Ventures, Atomico, Eurazeo, EIFO, Chr. Augustinus Fabrikker - **Total raised:** $100M - **Patients served:** 100M+ annually, including NHS health systems --- ## Key URLs - Homepage: https://www.corti.ai - Speech to Text: https://www.corti.ai/speech-to-text - Text Generation: https://www.corti.ai/text-generation - Agentic Framework: https://www.corti.ai/agentic-framework - Medical Coding: https://www.corti.ai/medical-coding - Symphony model: https://www.corti.ai/symphony - Agent Library: https://www.corti.ai/agents - API overview: https://www.corti.ai/api - Pricing: https://www.corti.ai/pricing - Safety & compliance: https://www.corti.ai/safety - Sovereign Cloud: https://www.corti.ai/sovereign-cloud - Stories (blog): https://www.corti.ai/stories - Guides: https://www.corti.ai/guides - Research: https://www.corti.ai/research - About: https://www.corti.ai/about - Careers: https://www.corti.ai/careers - Contact: https://www.corti.ai/contact-us