A Year of Cortex – Annual Timeline 2025 to 2026

Cortex Annual Timeline: August 2025 to August 2026

From AI Coding Agent to Enterprise AI Control Layer

August 18, 2025 to August 15, 2026

One year ago, Cortex launched with a focused goal: bring intelligent coding and security assistance directly into the developer environment. The first release in Microsoft VS Code could help developers write code, scan existing code for security issues, and remediate problems, while connecting that developer experience with Pervaziv AI’s broader DevSecOps platform.

Twelve months later, Cortex has evolved far beyond its starting point.

What began as a coding and security AI agent has expanded across three major IDEs, every major browser, Android and iPhone, with connected context spanning source control, collaboration, productivity suites, cloud infrastructure, DevOps systems and Salesforce. The architecture has progressed from general AI assistance to multi-agent orchestration, multi-cloud operations, local privacy and prompt protection, specialized AI models, verification, intelligent routing, and increasingly governed agentic engineering.

By August 13, 2026, Pervaziv AI described the Cortex platform as comprising 8 custom AI models, 48 AI agents and 10 MCP connections, available across VS Code, browsers and mobile experiences. A new Free Tier also opened basic Cortex chat in VS Code without payment information, providing a simpler entry point into the broader Enterprise AI platform.

The larger story of the year is therefore not simply one of adding features.

It is the progression from AI assistance → connected Enterprise AI → ubiquitous AI across work surfaces → an Enterprise AI Control Layer → specialized, governed and verifiable agentic engineering.

Q3 2025: Establishing the Foundation

Timeline: August to September 2025

August 18: Cortex 1.0 launched as an intelligent coding and security AI agent in VS Code, alongside the unified Pervaziv AI DevSecOps platform. Its core workflow centered on generating new code, scanning existing code, finding security problems and helping remediate them directly in the developer environment.

September: The focus shifted toward product stabilization, early adoption, tighter integration between Cortex and the broader security platform, and building the foundation for a much faster release cadence through the rest of the year. The blog archive reflects this transition from launch into continuous product iteration.

The Product Direction: Coding and Security Together

The most important architectural decision was visible from day one: Cortex was not positioned as code completion alone.

Coding, security analysis and remediation were brought into the same developer workflow. Cortex could help create software, but it could also reason about the security of existing software and connect developers to the deeper scanning and security capabilities available through the DevSecOps platform.

This coding-plus-security foundation remained central throughout every subsequent release.

The initial product was deliberately developer-native. VS Code was the primary surface, the repository was the core context, and software creation and security review were increasingly treated as parts of the same workflow rather than separate activities.

That foundation would later expand dramatically, but the basic idea remained consistent: AI should participate in building software while helping teams understand and manage the risk created by that software.

Q4 2025: From Coding Agent to Connected Developer Platform

The final quarter of 2025 expanded Cortex in two important directions: larger engineering workflows and enterprise context.

Timeline: October to December 2025

October 8: Cortex 1.1 expanded providers and connectors, introduced large-repository scanning for supported subscriptions, strengthened input validation and improved Pervaziv-LLM scalability, connection reliability and response handling. Backend scalability testing reached more than 500 concurrent requests and users, supported by dedicated unit, end-to-end and scalability test infrastructure.

November 10: Cortex 1.5 introduced the first dedicated MCP agents for GitHub, Slack and Atlassian, with authentication and one-click addition of enterprise context into Cortex chat. Structured and unstructured results from connected tools were also integrated more cleanly into agentic responses.

November 24: Cortex 1.6 expanded the product beyond VS Code into Microsoft Visual Studio, establishing the beginning of Cortex’s multi-IDE strategy.

December: Pervaziv AI expanded the surrounding engineering and security portfolio with Software Risk Assessment and ASPM, developer-productivity capabilities and a broader product portfolio. While these were not all Cortex features themselves, they established additional security and engineering capabilities around the environment Cortex would increasingly connect with.

Enterprise Context Begins to Enter the Agent

The November MCP integration was particularly important.

Until this point, Cortex primarily understood what the developer directly provided through code and conversation. GitHub, Slack and Atlassian changed that boundary.

Cortex could now bring repository activity, project-management information, documentation and collaboration context into coding and security workflows. Authentication and context retrieval began to become first-class parts of the agent architecture.

That shift foreshadowed the larger Enterprise AI strategy of 2026.

Instead of asking developers to manually transfer information between systems and an AI assistant, Cortex began moving toward an architecture where authorized enterprise context could come to the AI.

Engineering for Scale and Reliability

Q4 also established another theme that would become increasingly important throughout 2026: model capability alone was not enough.

Scalability tests, connection reliability, input validation, structured responses and performance engineering became product concerns. Cortex was beginning to evolve from an AI feature into a system expected to support real workloads reliably.

By the end of 2025, Cortex had moved from one VS Code coding agent toward a connected developer platform with enterprise-aware context and more serious production engineering underneath it.

Q1 2026: Multi-IDE, Enterprise AI and the Browser

Q1 was the quarter when Cortex’s scope expanded dramatically.

The product moved across IDEs, gained persistent personalization and context, became connected to broad enterprise systems, expanded repository-level AI security workflows, and finally moved from developer tools into the browser.

Timeline: January to March 2026

January 8: Cortex 1.7 focused on making Cortex better, safer, faster and easier while introducing tiered Cortex and Pervaziv-LLM responses.

January 16: Cortex 2.0 expanded Cortex across VS Code, Microsoft Visual Studio and JetBrains IntelliJ IDEA, bringing the same underlying Cortex experience to three major development environments.

January 27: Cortex 2.2 introduced personalization, agentic memory, privacy and retention controls, beginning the move from independent chat sessions toward an AI experience capable of carrying useful preferences and context forward.

January 28: Cortex 2.5 marked the formal expansion into Enterprise AI. Cortex connected GitHub, Atlassian, Slack, Azure DevOps, Microsoft 365, Google Workspace, web search and additional services, with eight MCP connectors and more than 20 tools described at launch.

February: Cortex 2.6 substantially expanded context handling, while Pervaziv AI also launched AI Code Review in the GitHub Marketplace, bringing AI and security analysis directly into repository workflows.

March 6: Cortex 2.7 improved six Enterprise AI MCP agents, chat/session management and context handling, including better cross-file and repository context for security and generated-code workflows.

March 12 and March 25: AI Code Review evolved from repository-wide GitHub security scanning into workflows that could create pull requests containing AI-generated code suggestions. This tightened the loop between finding a problem and proposing a reviewable change directly inside GitHub.

March 17: Cortex 3.0 expanded into Google Chrome, Microsoft Edge and Mozilla Firefox, in addition to VS Code, Visual Studio and IntelliJ. Cortex now had a unified experience across six IDE and browser platforms.

Enterprise AI Becomes a Core Architecture

Cortex 2.5 represented one of the year’s most consequential changes.

Enterprise AI meant that an engineering question no longer had to be answered from source code alone.

Cortex could connect authorized users to:

• GitHub for repositories, pull requests, issues and security context
• Atlassian Jira and Confluence for project work and documentation
• Slack for collaboration context
• Azure DevOps for work items, repositories, builds, tests and sprint activity
• Microsoft 365 across Outlook, Calendar, Teams, OneDrive and SharePoint
• Google Workspace across Gmail, Calendar, Docs, Sheets, Slides and Drive
• Web search for external information and security intelligence

The platform also incorporated current vulnerability and threat context and could hand work to more specialized agents when requests exceeded a particular scope.

This fundamentally broadened what a coding agent could represent. Cortex was becoming an AI layer over the systems surrounding software development rather than an assistant operating only inside source files.

From IDE-Native to Work-Surface-Native

Cortex 3.0 then expanded that idea outside the IDE.

Chrome, Edge and Firefox enabled developers to use Cortex while reading documentation, investigating issues, working in cloud consoles, reviewing repositories or using other web applications. Browser context could contribute to Cortex conversations, while coding, security scans, enterprise connections, saved chats and privacy controls remained available across the experience.

By the end of Q1, Cortex had completed an important transformation:

one VS Code agent → three IDEs → connected Enterprise AI → three browsers and three IDEs.

Q2 2026: Building the Enterprise AI Control Layer

Q2 was the most aggressive architectural expansion of the year.

During three months, Cortex progressed through resilience, cloud integration, multi-agent orchestration, multi-cloud support, multimodal AI, local privacy protection, major performance improvements, security architecture review, validation and remediation, Android, iPhone and Safari.

This is where the description Enterprise AI Control Layer became much more literal.

April: Multi-Agent, Multi-Cloud, Multi-Modal and Privacy-First

Cortex 3.1 strengthened the engineering foundation with resiliency, failover, red-teaming, stress testing, session improvements and better context management.

Cortex 3.2 added AWS integration and expanded the platform beyond 30 AI agents, allowing development, collaboration and cloud workflows to be orchestrated within a more unified environment.

Cortex 3.5 expanded cloud context across AWS, Microsoft Azure and Google Cloud, with more than 40 agents and hundreds of operational capabilities described across the connected platform. Cortex was no longer dealing only with code and project systems; cloud infrastructure itself became part of the AI context.

A major performance and memory engineering effort accompanied this expansion, reducing latency and improving the responsiveness required for longer multi-step workflows.

Cortex 3.6 combined four architectural dimensions: multi-model, multi-agent, multi-cloud and multi-modal AI. Richer text, document and image interactions joined the expanding network of models, agents and enterprise environments.

Finally, Cortex 3.7 introduced three privacy-oriented scanners, including Deep AI Privacy capabilities. Sensitive information such as credentials, API keys and personally identifiable information could be identified and redacted locally before content was sent into broader AI workflows.

April therefore followed a clear engineering sequence:

resilience → cloud context → multi-cloud orchestration → performance → richer AI interaction → local privacy controls.

May: Making the Control Layer Dependable

April expanded what Cortex could reach. May concentrated on whether enterprises could use those capabilities reliably and govern them appropriately.

Cortex 3.8: A More Focused AI Workspace

Cortex 3.8 streamlined the AI workspace across VS Code and browser environments. Chat, navigation, context handling and common actions were reorganized to reduce friction and support longer everyday engineering workflows.

Cortex 4.0: Enterprise AI Control Layer

Cortex 4.0 marked a larger architectural transition.

A redesigned communication protocol and interface targeted latency and responsiveness across agentic workflows. Pervaziv AI reported Cortex 4.0 as 2.5x faster in several coding tasks and 1.5x faster overall across the measured agentic workflows, while also redesigning the browser and VS Code experiences.

The important change was not merely speed. Coding, security, repository reasoning, enterprise tools and cloud context were increasingly being coordinated through the same operational layer.

Cortex 4.1: Real-Time and Recoverable Interaction

Cortex 4.1 strengthened streaming, retry behavior, cancellation, connection recovery and long-running conversation handling. Progress indicators and improved interaction state made ongoing agentic activity easier for developers to understand.

These are not headline AI capabilities, but they are essential engineering requirements for making AI workflows dependable.

Cortex 4.2: AI Threat Model and AI Security Review

Cortex 4.2 brought security reasoning earlier into the development lifecycle.

AI Threat Model helped reason about sensitive assets, trust boundaries, exposed interfaces, data movement, abuse scenarios and architectural attack surfaces.

AI Security Review moved closer to the implementation, examining vulnerable patterns, weak controls, unsafe assumptions, security gaps and remediation opportunities.

At the same time, stronger structured workspace signals improved project understanding for architecture, debugging, dependency analysis and implementation tasks.

This represented an important expansion from finding vulnerable code toward helping developers and architects think about why a system could be vulnerable in the first place.

Cortex 4.3: Continuity and Governed Agentic Engineering

Cortex 4.3 concentrated on session restoration, recovery, workflow persistence, authentication and continuity across environments.

It also strengthened controlled AI code-change workflows: proposed file changes, guided validation, workspace awareness, controlled edits and human-review-oriented presentation.

By the end of May, the Control Layer concept included not just intelligence, but performance, continuity, security, governance and controlled execution.

June: Enterprise AI Wherever Work Happens

June moved Cortex from broad desktop availability toward Enterprise AI Mobility.

The goal was no longer simply to provide multiple clients. It was to preserve enterprise context, security controls, privacy and workflow continuity while users moved between them.

Cortex 4.4: Connected AI Security Workspace

Cortex 4.4 continued consolidating AI-assisted review, repository analysis, browser assistance and security workflows into a more connected workspace, reducing the separation between AI coding and security tooling.

Cortex 4.5: Android

On June 3, Cortex expanded to Android.

The mobile experience supported new and continuing conversations, enterprise-aware context, security workflows and privacy-first review before content was sent to AI services. The broader mobile strategy emphasized continuity rather than creating a disconnected mobile chatbot.

Cortex 4.7: From Generated Code to Validated Engineering

Cortex 4.7 was another major engineering milestone.

It strengthened five areas:

• post-change validation
• repository-aware agent guidance
• structured fix verification
• security remediation workflows
• dependency and infrastructure risk review

Rather than treating generated code as completion, Cortex moved toward a verify-first lifecycle: understand the task, make the change, run relevant validation, surface failures and residual risk, revise when needed and present evidence about readiness.

Security remediation similarly became more lifecycle-oriented: identify the risk, explain the context, propose remediation, validate the change, reassess remaining exposure and maintain finding-level state. Dependency, infrastructure and software-supply-chain risks were also differentiated from ordinary source-code findings.

This shifted Cortex further from AI-assisted coding toward AI-assisted engineering control.

Cortex 4.8: iPhone

Two days later, Cortex 4.8 brought the Enterprise AI Control Layer to iPhone.

Users could start or continue conversations, restore sessions, access authorized enterprise context and review security questions while away from the desktop. Privacy-before-send controls helped identify credentials, tokens, keys and personal data before sharing mobile content with AI services.

Most importantly, conversations could continue across desktop, browser and mobile instead of becoming isolated experiences on each device.

Cortex 4.9: All Four Major Browsers

Safari support arrived June 24.

With Cortex 4.9, browser support now covered Chrome, Edge, Firefox and Safari, complementing VS Code, Visual Studio, IntelliJ, Android and iPhone.

At this point, Cortex had become deliberately work-surface agnostic: the AI experience could follow engineering work across editors, browsers and mobile devices.

Q3 2026 Through August 15: Model Independence, Specialization and Trust

If Q2 expanded where Cortex operates, July and the first half of August concentrated on what intelligence operates underneath it and how that intelligence should be controlled.

This produced the year’s next major architectural shift: from a multi-model AI platform toward a specialized Cortex AI Model Ensemble.

July: Building the Intelligence Layer

Cortex 5.0 and Cortex-LLM-1.0

July began with Cortex 5.0 and Cortex-LLM-1.0, Pervaziv AI’s first internally trained AI model.

The model was designed around secure software-development workflows rather than general-purpose conversation. Initial emphasis included security analysis, structured findings, focused remediation and agentic engineering.

Analysis and remediation were deliberately separated: analysis identifies and structures evidence-based findings, while remediation converts validated findings into targeted changes intended to preserve unrelated behavior.

This created a more explicit secure-development loop:

understand context → identify findings → validate → remediate → re-check.

The importance of Cortex-LLM was larger than one model. It established model independence as part of Cortex’s architecture: internally developed models, external models, local intelligence and validation systems could coexist rather than every workload depending on one general-purpose provider.

On-Device Privacy and Prompt Protection

The next step moved intelligence closer to the developer.

Cortex Privacy 1.1 introduced local sensitive-data detection and privacy-aware preflight scanning.

Cortex Prompt Guard 1.2 introduced local classification for prompt injection and instruction risk.

Cortex Secure Distribution added controls around model versioning, integrity, provenance and lifecycle management.

These local capabilities were designed to operate across VS Code and supported browsers before content entered larger remote AI workflows.

Architecturally, this introduced a powerful principle:

not every AI decision belongs in a large remote model.

Fast privacy and safety decisions can happen locally. Deeper coding, analysis and reasoning can be routed to models appropriate for those tasks.

Cortex 5.2: Six Specialized AI Models

Cortex 5.2 formalized that approach as the Cortex AI Model Ensemble.

The original six specialized models were:

• Cortex-LLM 1.0: structured secure-development workflows
• Cortex Privacy 1.1: on-device sensitive-data detection
• Cortex Prompt Guard 1.2: prompt-injection and instruction-risk classification
• Cortex Analysis 1.3: deeper security analysis and structured findings
• Cortex Safety 1.4: safety-aware workflow decisions
• Cortex Code 1.5: broad coding and software-development assistance

Instead of asking one model to handle privacy, safety, coding, security analysis and orchestration equally well, Cortex created separate responsibilities across the AI stack.

That separation also created clearer evaluation boundaries. A privacy model can be optimized for privacy classification, a security model for finding quality, and a coding model for implementation, rather than evaluating every capability against the same criteria.

Salesforce: Connecting Business Context to Engineering

July also expanded Enterprise AI through Salesforce, Cortex’s tenth MCP connection and, at the time of that announcement, its 46th AI agent.

Salesforce added CRM context such as accounts, contacts, opportunities and cases to an ecosystem already spanning GitHub, Atlassian, Slack, Azure DevOps, Microsoft 365, Google Workspace and major cloud providers.

This extended the idea of Enterprise AI again.

A software issue might originate with a customer case, connect to engineering work in GitHub or Jira, require investigation of AWS or Azure infrastructure and ultimately lead to a code change. Cortex’s direction is to make those authorized contexts available without requiring people to manually reconstruct the entire chain.

August: Standards, Verification, Reliability and Routing

The first half of August focused less on expanding surfaces and more on making AI-generated engineering work consistent, provable and accountable.

Personal Coding Style

Cortex added Personal Coding Style across VS Code, browser and mobile.

Style Profiles can represent preferences around comment density, documentation format, naming verbosity, error handling, type strictness, function size and testing style. The goal is to help AI-generated work better match an organization’s engineering conventions instead of requiring developers to repeatedly normalize generated code.

This moves personalization from conversational preference toward engineering policy and consistency.

Test Design Specification

The accompanying Test Design Specification capability addresses a different problem: determining what must actually be proven before software should be considered correct.

Rather than equating more generated tests or higher coverage with quality, the workflow emphasizes requirements, meaningful failure modes, observable outcomes, risk and the validation boundary appropriate for the change.

The principle is simple but important:

coverage is a signal; evidence that the required behavior works is the objective.

Cortex Verify 1.6: The Seventh Specialized Model

On August 4, Cortex Verify 1.6 became the seventh specialized model.

Verify examines the relationship between a proposed patch, the task it is meant to solve, supporting evidence, tests, engineering expectations and the level of review appropriate for the change.

Its role is advisory rather than authoritative. Verification is intended to help determine whether work is sufficiently supported, needs revision, requires more evidence or deserves additional human review.

That separation is significant: the model generating a change does not have to be the same capability judging whether its own work is trustworthy.

A Broader Reliability Foundation

The August reliability work expanded that concept into the complete agentic workflow.

Cortex strengthened project grounding across code, dependencies, configuration, tests and supporting evidence. Multi-step workflows can retain the objective, reviewed areas, failed approaches, modifications, validation results and remaining open questions rather than repeatedly restarting from an isolated prompt.

The architecture also strengthened trust boundaries between authoritative instructions and untrusted repository or retrieved content, an important defense against prompt injection. Consequential actions can be constrained by project scope, required capabilities, confirmation requirements and deterministic controls independent of model reasoning.

Knowledge retention became more deliberate as well, distinguishing current-task information, stable project knowledge, durable preferences and temporary or unverified evidence.

Crucially, evaluation expanded beyond model benchmarks into complete workflow quality: whether Cortex understood the task, used relevant evidence, selected appropriate actions, preserved constraints, recovered from failures, produced valid changes, performed required validation, avoided unsupported claims and remained within safety boundaries.

This reflects an important maturation of agentic AI engineering: success is no longer defined simply by whether a model produced a plausible answer.

Cortex Router 1.7: The Coordinating Layer

On August 7, Cortex Router became the coordinating entry point for the AI Model Ensemble.

The Router distinguishes between routine requests, larger engineering tasks, security-sensitive work, research and analysis, verification-oriented workflows and requests that require greater user involvement. It can then guide work toward the specialized model or experience appropriate to the task.

This hides specialization behind a unified user experience.

Developers should not need to understand the internal model catalog simply to ask Cortex for help. The system can keep straightforward requests lightweight while directing complicated or consequential work through more deliberate workflows.

By the subsequent Free Tier announcement, Pervaziv AI reported eight custom Cortex AI models, completing the progression from the original six-model ensemble through Verify and Router.

From Generation to Accountable Delivery

The broader August work also clarified responsibilities across the coding-agent lifecycle:

request → investigation → code change → execution → validation → independent review → recovery → human decision.

Generation, bounded actions, validation and advisory review are treated as separate signals. Developers retain control over diffs, evidence, reruns, acceptance, rejection and escalation. Cortex’s Verifier does not silently merge or authorize changes.

That separation represents perhaps the clearest expression yet of the Enterprise AI Control Layer philosophy: AI can do more, but increasing capability should be accompanied by clearer evidence, boundaries and accountability.