LocalIntake Review
A local-LLM clinical intake agent that runs on Ollama and keeps patient data on the device.
Each review covers deployability, value versus commercial alternatives, and privacy model. Tools that can run locally were started and exercised; mobile or backend-dependent tools were assessed from published builds, source code, and deploy guides. Ratings reflect what we were able to verify.
LocalIntake — Open Source Local-LLM Clinical Intake
A local-LLM clinical intake agent that runs on Ollama and keeps patient data on the device. Last updated: 2026-06-21.
One-sentence verdict: The strongest privacy model in this batch: a local-LLM intake agent with structured output and PDF reporting, held back only by the need to download and run a 7 GB local model.
What the System Is
LocalIntake (published as local-health-intake-ai) is a Python Streamlit app that interviews a patient before a clinical visit. The deploy guide reports the Streamlit frontend was started on port 5019, but Ollama was not installed in the environment, so the LLM backend could not connect.
The agent extracts and validates the following structured fields through conversation:
- Chief complaint in the patient's own words
- Duration of the symptom
- Severity score (1–10)
- Additional symptoms
- Allergies and current medications
- Red flags such as chest pain or difficulty breathing
- Suggested triage urgency (Routine, Urgent, Emergency) with reasoning
Output is enforced by a Pydantic schema, and the app can generate a PDF report.
| Key data | |
|---|---|
| Category | Clinical Intake |
| Language | Python / Streamlit |
| License | MIT |
| Self-hosted | Yes |
| AI provider | Ollama (local) — mistral-nemo |
| Output | Pydantic-validated structured intake + PDF report |
| Data residency | 100% local inference |
How to Install and Deploy
cd /data2/docker/going-global/repos/local-health-intake-ai
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt
# In one terminal:
ollama pull mistral-nemo
ollama serve
# In another terminal:
streamlit run app.py --server.port 5019 --server.headless true
Then open http://localhost:5019.
The README defaults to port 8501; the deploy guide used 5019.
How to Test
The documented test flow is:
- Ensure Ollama is running and
mistral-nemois pulled. - Open
http://localhost:5019. - Start an intake conversation and answer the symptom questions.
- Verify the app extracts structured fields and assigns a triage urgency.
- Generate the PDF report and confirm it contains the collected information.
The deploy guide notes that without Ollama the app shows a ConnectionError: Failed to connect to Ollama.
Privacy & Compliance
LocalIntake is not HIPAA compliant, but its privacy architecture is the best in this batch. All inference runs locally via Ollama; patient data does not leave the device. There is no authentication, encryption, or audit trail, so it is still unsuitable for regulated PHI without additional hardening.
LocalIntake vs Commercial Intake Systems
| Dimension | LocalIntake | Commercial Intake (e.g., Notable Health, Mend, Phreesia) |
|---|---|---|
| Cost | Free / self-hosted | Subscription per provider |
| Data model | Local inference; no third-party transmission | Cloud-hosted |
| Offline use | Yes, after model download | No |
| Clinical validation | None | Medically reviewed, EHR-integrated |
| Setup effort | Python venv + Ollama + 7 GB model | Vendor-managed |
| Report format | EHR-compatible structured data |
Who Should Use It
- Teams that need a fully local intake prototype to avoid cloud PHI exposure.
- Developers evaluating Ollama-based medical chat workflows.
- Privacy-conscious pilots where data must stay on device.
Who Shouldn't Use It
- Organizations needing HIPAA-compliant intake without additional engineering.
- Environments where downloading and serving a 7 GB model is impractical.
- Workflows requiring EHR integration, scheduling, or clinician dashboards.
FAQ
Does LocalIntake send data to the cloud?
No. The architecture is designed for 100% local inference through Ollama. No chat data is transmitted to third-party LLM APIs.
What model does it require?
The README and deploy guide use mistral-nemo, pulled through Ollama. The model is roughly 7 GB.
Was the LLM backend running in the deploy environment?
No. Ollama was not installed, so the Streamlit UI started but could not connect to a model.
Verdict
LocalIntake is a compelling proof of concept for on-device clinical intake. The combination of conversational input, Pydantic validation, triage urgency, and PDF reporting is practical, and the local-only inference removes the cloud-privacy risk. It needs hardening for real clinical use, but the architecture is sound.
Ratings: Deployability 3/5 · Value vs Commercial 3/5 · Privacy Compliance 5/5