Clone the repository

git clone https://github.com/clinia/context-engine-studio.git
cd context-engine-studio

Configure environment variables

Create a .env.local file with your workspace URL, OAuth credentials, and Anthropic API key:

cat > .env.local <<'EOF'
CLINIA_CONTEXT_ENGINE_API_URL=https://<workspace-id>.w.clinia.cloud
CLINIA_CONTEXT_ENGINE_OAUTH_CLIENT_ID=<your-client-id>
CLINIA_CONTEXT_ENGINE_OAUTH_CLIENT_SECRET=<your-client-secret>
ANTHROPIC_API_KEY=<your-anthropic-api-key>
EOF

Replace each placeholder with your actual values. See Manage Credentials if you need to create new OAuth credentials.

Install and start

pnpm install
pnpm dev

Open http://localhost:3000. If no patients exist in the workspace yet, Studio sends you straight to the onboarding screen.

Ingest the synthetic dataset

Download the Jeanne Tremblay dataset, which contains a FHIR R4 bundle and C-CDA documents for a fully synthetic 72-year-old patient with a decade of clinical history.

Unzip it and drop the folder onto the Studio onboarding dropzone. The engine ingests both FHIR and CDA sources together, running cross-source entity resolution across them.

See the Synthetic Patient Dataset page for a full portrait of Jeanne and what the dataset contains.

Browse the virtual file system

Once the ingest completes, the Studio shows the patient overview. Expand the virtual file system tree in the sidebar to navigate the patient story as paths:

• /conditions/active/ lists every active condition the engine resolved.
• /medications/active/ shows the current medication regimen.
• /encounters/ contains every encounter, ordered in time.

Select any node to view its content. Each file is available as a Narrative, Compact, or Structured view.

Read a condition story

Navigate to /conditions/active/ and select a condition (for example, chronic_obstructive_lung_disease). Open the Narrative tab to read the condition story: a pre-assembled longitudinal narrative covering onset, active treatments, monitoring labs, complications, and contributing factors, all sourced from the resolved graph.

This is the same content an agent retrieves when it calls the read_patient MCP tool. See Condition Stories for how stories are assembled.

Chat with the agent

Open the chat panel and ask a clinical question that requires reasoning across the graph. For example:

Given the recent acute COPD exacerbation, is the current inhaler regimen appropriate and is bone health being monitored?

The agent browses the VFS, reads condition stories, and synthesizes an answer with citations pointing back to the files it used. You can follow the tool calls in the chat to see exactly which paths the agent navigated.

Chat threads are saved per patient in a local SQLite file (.data/studio-chats.db by default) and reopen where you left them.