Cortex: from ticket to reviewed pull request
A developer drags a ticket across a Kanban board and a chain of five AI agents plans, codes, reviews, tests and prepares the deployment, all running on the developer's own machine.
- Industry
- AI and developer tools
- Services
- Product build, AI agent pipeline, real-time dashboard
- Stack
- Next.js 15, FastAPI, PostgreSQL with pgvector, Socket.IO
- Runs on
- Local runner on the developer's machine
The problem
AI coding assistants handle a single prompt well, but shipping a feature also means planning, review, testing and deployment. Teams wanted that whole loop automated without giving an opaque tool access to their codebase, secrets or production environment.
How we approached it
We split the work across five agents, each with one job and a checkpoint before the next starts. The runner executes locally, so code stays on the developer's machine. A scope guard keeps every agent inside the files a ticket allows, and a semantic Project Brain gives the agents context from the codebase and past decisions.
What we built
Five-agent pipeline
Planner, Coder, Reviewer, QA and DevOps agents pass work along, each checking the previous step before anything moves on.
Live Kanban board
Move a ticket from backlog to done and see commits, runner status and agent progress update in real time.
Scope guard
Agents can only touch the files declared in a ticket and can never read secrets or env files. Checked before every commit.
Local runner CLI
A small runner connects any git repository and runs on the developer's machine using their existing AI coding subscription.
Meeting notes to tickets
Paste a transcript and Cortex turns action items into tickets, flags blockers and records the decisions made.
Project Brain
Semantic search across the codebase and past decisions, with answers that point to the exact file and line.
Tech stack
- Next.js 15
- TypeScript
- Tailwind CSS
- FastAPI
- Python
- PostgreSQL
- pgvector
- Valkey
- Socket.IO
- Celery
- Docker
- sentence-transformers
Results
Developers move tickets across a board and get back reviewed, tested changes ready to merge, with a record of what each agent did and why. Routine engineering work takes less time, the guardrails keep the automation in check, and the team spends its attention on decisions that need a person.
What does Cortex do?
Cortex is an AI engineering platform that turns a ticket on a board into a reviewed, tested pull request. A chain of five specialized agents plans the work, writes the code, reviews it, runs QA, and prepares the deployment, with a live Kanban view of everything as it happens.
Is it safe to let AI agents work in a codebase?
Cortex enforces a scope guard so agents can only modify the files declared in a ticket and can never read secrets or environment files. The runner executes locally on the developer's own machine, so source code never leaves their environment.
How does Cortex understand an existing codebase?
A semantic Project Brain indexes the entire codebase along with past meeting decisions, so agents answer questions and plan changes with real context, citing the exact file and line an answer came from.
Want an agent pipeline for your engineering team?
We can design a plan, build, review and ship workflow around your repositories, with guardrails you control.