Research & Technology

The science behind Convai Agent OS

Our platform leverages cutting-edge research in on-device LLMs, agentic AI architectures, and sandboxed execution environments to deliver healthcare-grade AI.

Local Multimodal Agentic LLM — On-Device Intelligence

State-of-the-art open-weight model

Convai Agent OS ships with a local multimodal agentic LLM, a state-of-the-art open-weight model optimized for on-device deployment. It achieves near-GPT-4 quality on reasoning, summarization, and instruction-following benchmarks while running entirely locally.

We use LiteRT (formerly TensorFlow Lite Runtime) for inference, enabling GPU-accelerated execution via Vulkan on NVIDIA and AMD hardware. The model runs in 4-bit GGUF quantization, requiring only 4-6GB VRAM for full-speed inference.

4.6B
Parameters
Optimal size for local deployment
~35
Tokens/sec
On RTX 3060 (6GB VRAM)
4-bit
Quantization
GGUF Q4_K_M format

Bring Your Own Model (BYOM)

Specialized medical models for specialized needs

While the default local multimodal agentic LLM is exceptional for general-purpose healthcare tasks, some clinical workflows benefit from domain-specialized models. Convai Agent OS supports hot-swapping to any GGUF-compatible open-source model directly from the desktop UI.

ModelParametersSpecialtySource
Local Multimodal Agentic LLMGeneral-purpose (default)Bundled with Convai Agent OS
Meditron7B / 70BClinical decision supportEPFL
ClinicalCamel70BMedical Q&A and reasoningWanglab
BioMistral7BBiomedical literature analysisBioMistral Project
Med-PaLM (open)7BMedical exam-level reasoningCommunity fine-tune

Agentic AI Architecture

Autonomous multi-tool task execution

Convai Agent OS is not a chatbot — it's an autonomous agent. The LLM has access to a suite of tools and can chain them together to complete complex, multi-step healthcare workflows without human intervention.

Document Generator

Creates PDF discharge summaries, DOCX referral letters, XLSX tracking sheets from natural language instructions

Patient Memory (RAG)

Semantic vector search across all uploaded patient records. Ask questions about any patient's history.

Web Research

Multi-round deep web search for drug interactions, treatment protocols, and latest medical literature

Communication Gateway

Sends messages, documents, and reports via Telegram, WhatsApp, or Email automatically

File Management

Creates, organizes, and searches patient folders. Converts handwritten notes to structured PDFs.

Scheduling Engine

Cron-based appointment reminders and follow-up notifications via messaging channels

Example: Autonomous Discharge Summary
Doctor: "Create a discharge summary for patient Ravi Kumar, Room 402"
Agent → [Tool: Patient Memory] Searching records for Ravi Kumar...
Agent → [Tool: Patient Memory] Found: admission notes, 3 lab reports, 2 prescriptions
Agent → [Tool: Document Generator] Generating discharge summary PDF...
Agent → [Tool: Email Gateway] Sending to ravi.kumar@email.com...
✓ Discharge summary created and emailed to patient. PDF saved to /patients/ravi-kumar/

Sandboxed Custom OS

Hardware-level isolation via QEMU virtualization

Unlike generic AI tools that execute directly on your operating system with full access to your files, Convai Agent OS runs its entire AI stack inside a purpose-built Linux distribution within a QEMU virtual machine. The AI literally cannot access your host computer — it's physically isolated at the hardware emulation level.

Why this matters vs. OpenClaw & other agents

❌ Generic agents (OpenClaw, etc.)

  • • Execute code directly on host OS
  • • Full filesystem and network access
  • • Can install packages, modify system files
  • • No healthcare-specific controls

✅ Convai Agent OS

  • • Runs inside isolated QEMU VM
  • • Zero access to host filesystem
  • • Strict port-forwarded networking
  • • HIPAA audit logging built-in

Ready to experience healthcare AI done right?

Local LLM. Sandboxed OS. HIPAA-compliant. No cloud. No compromise.