/ overview
Inspect the claim behind the profile.
CortexOS is a desktop intelligence system for researching companies and organizing business signals. The important engineering idea is the boundary between generated claims and source evidence.
A React and TypeScript interface runs inside Tauri. Rust handles application capabilities, agent workflows, verification, and local persistence through SQLite.
/ problem
Plausible text is a weak data contract.
A company profile can look convincing even when a title, technology, or metric has no source. Downstream workflows need a way to inspect those details rather than treating generated prose as authoritative.
CortexOS separates claim extraction from verification and exposes evidence in the product interface.
/ architecture
Research first. Verify before presenting.
Agent workflows gather company context and extract candidate claims. A deterministic Rust path compares claims against retrieved source text using exact, normalized, and fuzzy matching.
Claims and their source context flow into the local application, where users can inspect evidence. SQLite provides local persistence, while Tauri connects the frontend to the Rust application core.
- Specialized research workflows are coordinated by an orchestrator.
- Verification compares against source text rather than model confidence.
- The frontend exposes evidence alongside generated intelligence.
/ Engineering decisions
A verifier has a precise, limited job.
Deterministic matching makes a verification decision reproducible and easier to inspect. Text appearing in a source does not establish that the source itself is correct, and fuzzy similarity does not establish semantic truth.
The interface keeps source evidence available alongside each result so the user can inspect the basis for a match.
Local persistence keeps the desktop application’s state near the user and adds practical concerns around migrations, background work, and API credentials.
/ challenges
Keep the UI connected to the evidence.
An agent system has multiple stages with different failure modes: fetching, extraction, verification, storage, and presentation. The interface needs to reveal the source behind a result while the runtime handles long-running work.
The codebase separates Rust application modules and frontend views so verification and orchestration can evolve without turning every interface component into a model integration.
/ results
A working local intelligence application.
The public repository contains the Tauri application, React interface, Rust agent and verification modules, and SQLite-backed data layer.
The implementation connects research, source matching, and evidence inspection in one desktop workflow. Source reliability and matching errors remain distinct evaluation questions.
/ learnings
Trust should be explorable.
A result is more useful when its evidence is one interaction away. The product’s evidence surface is part of its architecture, alongside the retrieval and verification logic.
A useful next evaluation would separate extraction quality, source reliability, matching errors, and downstream decision quality rather than combining them into one unqualified score.
Source record
Sources & technical context
Architecture summaries are based on the linked documentation and source. Current implementation details and limits live in the repository.
- Product overview and architectureReviewed
- Rust application coreReviewed