Architecture and security.Layers with clear boundaries.
Codira's architecture separates the models that perform tasks from the layers that govern them: orchestration, specialist agents, deterministic controls and human authority, with each product built on top.
Foundation models come from third-party providers and change quickly. Codira's own engineering is the set of layers above them: orchestration that plans and sequences work, agents with defined roles, rules in code that check results, and approval steps that keep people in charge. Keeping those layers independent of any one model is what lets the safeguards stay in place when the models change.
Security follows the same principle of explicit boundaries. Each product limits what its AI can reach and do, separates one customer's data from another's, keeps an audit trail, and treats content from documents and the web as data rather than instructions. This page describes those principles; it does not publish internal configuration.
Security principles.The boundaries behind each product.
Separation of Layers
Customer Isolation
Least Authority for AI
Untrusted Content
Strong Sign-In
Audit Trails
Model interoperability.Change the model, keep the safeguards.
Because orchestration is separate from the models, a task can be routed to the model best suited to it, and a new model can be adopted without redesigning the workflow around it.
The rules, approvals and audit trail sit in Codira's layers, so they carry on unchanged when the model underneath does.
In our products.Boundaries, product by product.
Each product applies the principles to its own data and risks.
AI file access is confined to the project folder, every command is classed as safe, needing approval or blocked, and keys are kept in the macOS Keychain.
The productRow-level security keeps every organization's data apart in the database, uploads are scanned for malware, and an append-only audit log includes staff actions.
The productPer-case roles and ethical walls also exclude screened documents from AI, privilege is decided by counsel, and an append-only audit log records activity.
The productWhat it does not do.What this page does not claim.
Security statements are kept to what the products do today.
No certification such as SOC 2 is claimed for any product.
Internal infrastructure, configurations and source code are not published here.
Security practices differ by product; each product's own security page has the specifics.
The rest of the architecture.Explore the other capabilities.
An architecture builtto outlast any single model.
Clear layers and clear boundaries let Codira adopt better models while keeping the controls professionals rely on.