8 June 2026Three modes · same product
/ security

Three modes. You draw the boundary line.

Intelectra runs in three deployment modes. The retrieval and verification pipeline is identical across all three. What changes is where the LLM runs and where the services that orchestrate it live.

Architecture
MODE · 01OpenLLMTHIRD-PARTYAgent orchestrationINTELECTRA · EUROPERAG + paper databaseINTELECTRA · EUROPEPubMed retrieverINTELECTRA · EUROPEBEST FORPublic research · fastest setupMODE · 02HybridLLMYOUR GPUAgent orchestrationINTELECTRA · EUROPERAG + paper databaseINTELECTRA · EUROPEPubMed retrieverINTELECTRA · EUROPEBEST FORNDA-sensitive prompts · neutral retrievalMODE · 03ClosedLLMYOUR GPUAgent orchestrationYOUR SERVERSRAG + paper databaseYOUR SERVERSPubMed retrieverYOUR SERVERSBEST FORAir-gapped labs · pharma R&D · clinical
§ 01Mode · Open

Open. Third-party LLM · Intelectra services hosted in Europe.

WHAT LEAVES YOUR MACHINE
  • Your prompt and any text you include reaches the third-party LLM vendor.
  • Your PubMed search terms reach NCBI as queries.
WHAT STAYS
  • Documents in your desktop workspace stay on your machine.
  • The bibliography state lives in your local workspace.
BEST FOR
  • Public research, literature reviews, teaching. You need fastest setup and don't want to host anything yourself.
§ 02Mode · Hybrid

Hybrid. Self-hosted LLM · Intelectra services hosted in Europe.

WHAT LEAVES YOUR MACHINE
  • Your prompts never reach a third-party LLM vendor. They go to the LLM you host on your own GPU.
  • Retrieval queries (which PMIDs and which chunks) reach the Intelectra services in Europe; the prompt body does not.
  • Your PubMed search terms reach NCBI as queries.
WHAT STAYS
  • Your manuscript text, your prompts, and the LLM that reasons about them stay on your hardware.
  • All raw documents in your workspace stay on your machine.
BEST FOR
  • NDA-sensitive prompts, sponsor-confidential drafts, lab-internal hypotheses. You want neutral retrieval but you control the reasoning model.
§ 03Mode · Closed

Closed. Self-hosted LLM · self-hosted Intelectra services.

WHAT LEAVES YOUR MACHINE
  • Optional PubMed queries to NCBI if you choose to enable them; otherwise the system is fully offline.
WHAT STAYS
  • Every component runs on your infrastructure: LLM, agent orchestration, retrieval, paper database, embeddings.
  • Documents, prompts, retrieval queries, and reasoning logs never leave your network.
BEST FOR
  • Air-gapped labs, pharma R&D, clinical research with unpublished trial data, regulated environments. Maximum control, all components disk-resident.
§ 04Threat model

What we defend against, and what we don't.

Honest threat model. What Intelectra structurally prevents, what we're explicit about, and where the boundary actually ends.

DEFENDED · STRUCTURAL
  • Hallucinated citations reaching your document: quote verification gates persistence; nothing unverified is saved.
  • Documents leaving your machine in Closed mode: no component has an external network path.
  • Prompts leaving the LLM you control in Hybrid and Closed mode: the third-party vendor is removed from the loop.
  • Tool calls without provenance: every call is logged alongside the document and the verified quote it produced.
EXPLICIT TRADE-OFF
  • In Open mode, the third-party LLM sees the prompt, including any context you include. It's an opt-in path, declared at call time.
  • PubMed queries reach NCBI's servers: your search terms are not private.
  • Retrieval queries in Hybrid mode reach the Intelectra services in Europe: the prompt body does not, but the PMIDs and chunk IDs you request do.
OUT OF SCOPE
  • Endpoint compromise: if the host machine is compromised, your data is reachable via the filesystem.
  • Disk encryption: handled by your OS, not us.
  • Multi-tenant isolation: Intelectra is a single-tenant desktop app per researcher.
§ 05Hosted service

Using the hosted public sandbox or Open mode? The Trustpage names every sub-processor that touches data, where the service runs, and what is and isn't encrypted.