Frequently Asked Questions

Cortex is an institutional knowledge platform for AI-assisted engineering teams. It captures the decisions, discoveries, patterns, and domain knowledge generated during AI sessions, persists them, and surfaces relevant insights in future sessions. Every session your team runs makes the next one smarter.

AI-assisted engineering sessions generate valuable knowledge (API quirks, architecture decisions, debugging root causes, domain context) that disappears when the context window clears. Across a team running dozens of sessions per day, this means repeated debugging, lost decisions, and engineers starting from scratch on problems that have already been solved. Cortex captures that knowledge automatically so it compounds instead of evaporating.

Documentation tools require manual entry and discipline that doesn't scale. Cortex operates at the layer where knowledge is actually generated: inside the AI session itself. Capture is automated, quality-filtered, and classified. Engineers don't need to context-switch to a separate tool, write a doc, or tag anything manually. Cortex captures the knowledge that would never make it into a wiki.

Cortex classifies insights into six types: Decisions (choices between alternatives with reasoning), Discoveries (API quirks, debugging root causes, corrections), Domain Knowledge (business rules, product context), Patterns (conventions, workflows, style preferences), Environment (setup, config, deployment details), and Performance (bottlenecks, benchmarks, scaling limits). Each insight carries a confidence score and metadata for precise retrieval.

No. Cortex installs as a plugin and operates at session boundaries via hooks. Engineers work exactly as they do today. Capture is triggered at session end. In auto mode, insights are extracted without intervention. In the default mode, Cortex prompts the engineer to review and confirm. Relevant past insights are surfaced automatically when a new session starts. The core value is delivered without workflow changes.

Cortex uses a three-tier privacy model. Personal insights (individual preferences) stay completely private. Project insights (repo-specific knowledge) are shared only within that project's team. Team insights are explicitly sanitized and generalized before crossing project boundaries. Proprietary details, client names, and specific implementation details are stripped. The default scope is project-level, and every cross-boundary insight goes through sanitization validation.

Your knowledge lives in a dedicated, enterprise-grade PostgreSQL database on your own Google Cloud project. You own the database, the data, and the infrastructure. We manage provisioning and operations, but nothing is hosted on VGV systems or shared with AI vendors. All data is encrypted at rest and in transit. We support configurable data residency for organizations with geographic compliance requirements.

Enterprise deployments include SSO integration, role-based access controls, and audit logging. You can configure which teams have access to which knowledge scopes, set custom retention policies, and maintain a full audit trail of what was captured, queried, and by whom.

Insight text is sent to Google's embedding API to generate vector representations for semantic search. All other data stays within your database instance. For self-hosted deployments, the embedding API calls originate from your own cloud project. No insight content is used for model training, and no data is shared with any party beyond what's required for embedding generation.

Yes. Cortex supports self-hosted deployments on your own Google Cloud infrastructure. All managed components (database, API server, embeddings) run within your cloud project, so data never leaves your environment. We provide the schema, configuration, and deployment tooling. You control the infrastructure.

Cortex currently integrates with Claude Code via its plugin and hooks system. The architecture supports any AI assistant that implements MCP (Model Context Protocol) or similar extension mechanisms, so additional integrations are straightforward to add.

We provision a dedicated database instance and API server on your Google Cloud project, then configure the plugin for your team. Your engineers install Cortex, authenticate, and start working. For organizations that prefer to run setup themselves, we provide infrastructure-as-code scripts and onboarding support. Either way, the infrastructure lives in your environment.

Cortex uses hybrid search combining vector similarity for semantic understanding with full-text ranking for exact term matching. Results are weighted by confidence scores that decay over time based on domain-specific rates, plus reinforcement signals from past retrieval. Queries work whether you use the exact terminology from the original insight or describe the concept in your own words.

Cortex runs on a PostgreSQL database with vector extensions, an API server, and inline embedding generation, all deployed on your cloud project. We handle provisioning and day-to-day operations. The infrastructure cost is a single database instance plus API calls for embeddings, which is modest relative to the engineering time it recovers.

The knowledge base starts delivering value from the first captured insight. Within the first week of deployment, engineers typically see relevant past insights surfacing during their sessions. The value compounds over time: as the knowledge base grows, engineers hit fewer dead ends and spend less time re-solving known problems.

Cortex tracks several quantitative metrics: number of insights captured per session, retrieval hit rates (how often surfaced insights are relevant), reduction in repeated issue debugging, and engineer onboarding time. The most direct signal: every discovery insight that prevents a repeated debugging session is hours of engineering time recovered. Multiply by the number of engineers and sessions across your organization.

Cortex is designed for teams that are already using AI-assisted engineering as part of their daily workflow. If your team is in early adoption, Cortex can accelerate that process by ensuring early learnings are captured and available to the broader team, reducing the trial-and-error phase for each new engineer. We're happy to discuss fit for teams at different adoption stages.