Every architectural decision. Every tradeoff. Every signal from every surface your team works on — captured automatically, threaded intelligently, and surfaced the moment anyone needs it. Without changing how anyone works.
Redis selected over Memcached for rate limiting
Persistence and pub/sub support required at scale
Proposed · 84% confidence
Project name detected in Google Doc
We should standardise error response format across all API endpoints
Added to thread
Kortex CI scan: 1 critical violation
Input validation bypassed before business logic. Contradicts affirmed decision from 14 weeks ago.
Build blocked
@kortex why do we use a monorepo structure?
Kortex replied in thread · 3 sources cited
Answered
processPayment — nonconformity flagged
Error handling missing. Affirmed pattern: all payment functions must handle timeout explicitly.
Warning · inline
INFRA-091 completed: migrate logging to structured JSON
Across all services — completed and linked to prior architecture decision
Added to thread (2 signals)
Removed custom OAuth — replaced with provider SDK
Security review flagged implementation complexity risk
Affirmed · 93% · Thread: OAuth strategy (5 signals)
Project name detected in Outlook email
The reason we chose PostgreSQL over MongoDB was consistency requirements for transactions
Proposed · Saved to memory
Redis selected over Memcached for rate limiting
Persistence and pub/sub support required at scale
Proposed · 84% confidence
Project name detected in Google Doc
We should standardise error response format across all API endpoints
Added to thread
Kortex CI scan: 1 critical violation
Input validation bypassed before business logic. Contradicts affirmed decision from 14 weeks ago.
Build blocked
@kortex why do we use a monorepo structure?
Kortex replied in thread · 3 sources cited
Answered
processPayment — nonconformity flagged
Error handling missing. Affirmed pattern: all payment functions must handle timeout explicitly.
Warning · inline
INFRA-091 completed: migrate logging to structured JSON
Across all services — completed and linked to prior architecture decision
Added to thread (2 signals)
Removed custom OAuth — replaced with provider SDK
Security review flagged implementation complexity risk
Affirmed · 93% · Thread: OAuth strategy (5 signals)
Project name detected in Outlook email
The reason we chose PostgreSQL over MongoDB was consistency requirements for transactions
Proposed · Saved to memory
Every signal compounds. Every thread grows stronger. The intelligence never stops building.
Kortex doesn't stop at answering questions. Plan turns a request into a real, grounded implementation brief — explored against your actual codebase, not a guess — before a single line changes. Build then executes that brief end to end.
Ask
Understand the system before you touch it. Answers grounded in project memory, with sources you can inspect.
Plan
Explores your real codebase, asks a genuine clarifying question when something is truly ambiguous, and hands back a grounded brief with an honest scope estimate — download it, hand it off, or send it straight to Build.
Build
Executes Plan's brief end to end: writes the code, runs the tests, fixes its own mistakes, checks its own work — then remembers what it built.
Your team already decided how auth should work, how errors should be handled, which library you standardized on. Kortex remembers all of it and checks every PR against it, before it merges.
Kortex
This reintroduces the custom OAuth flow removed in #847 after a security review flagged implementation complexity risk.
Conflicts with affirmed decision - OAuth strategy thread - 93% confidence
Continuous scanning for hardcoded secrets, SQL injection, insecure crypto, missing auth checks, and known dependency vulnerabilities. Every week, automatically. Every Build, before it is handed back.
Hardcoded API key in config/payments.ts
Detected during Kortex Build - 2m ago
Missing auth check on /admin/export endpoint
Weekly scan - core-api
Dependency: lodash 4.17.15 - known prototype pollution
Weekly scan - dependency audit
Kortex does not change how your engineers work. It changes what gets lost when they do.
A senior engineer leaves on Friday. On Monday, a new engineer joins. They spend the next six weeks asking questions nobody can fully answer, reverse-engineering decisions nobody remembers making, and occasionally repeating mistakes that were already made and corrected two years ago.
~$40,000
in lost productivity. Every time a senior engineer leaves.
A senior engineer leaves on Friday. On Monday, a new engineer joins. Kortex has already generated their onboarding guide — every architectural decision, every pattern to follow, every mistake to avoid — and emailed it to them before their first standup.
Productive in 2 days.
The onboarding guide generates itself. No effort required.
This is not a documentation tool. Documentation requires intention.
Kortex requires nothing. It just watches.
How well understood is this repository, really? Coverage, confidence, and freshness, in one score.
What would you lose if a specific engineer left tomorrow?
Which parts of the codebase are understood, stale, drifting, or still dark.
What your team decided, and how those decisions evolved, in one exportable record.
A new engineer first-week context, generated automatically.
Kortex keeps a living .kortex/context.md file in your workspace, read automatically by Claude Code, Cursor, Copilot, and Windsurf, with zero configuration. Agents can also query Kortex directly through MCP for anything the context file does not cover.
"We have multiple projects running in parallel and Kortex has become the connective tissue between them. Our engineers no longer ask each other why the codebase works a certain way — they ask Kortex. The quality of our decisions has improved because we can finally see the history of every one of them."
Head of Software Engineering
Dragnet Solutions
A decision is made in a Slack message. Kortex captures it. Three weeks later a developer commits code that implements that decision — Kortex links them into a thread. A PR is opened — the CI/CD action checks the code against the decision. A product manager writes a spec in Notion — the browser extension surfaces the thread. A new engineer asks @kortex in Slack — the answer cites all four signals. One intelligence thread. Five surfaces. Zero extra effort.
Intelligence at every function. Q&A without leaving your editor. Nonconformities flagged inline.
Every PR checked against every decision your team has affirmed. Violations flagged before merge.
@kortex answers questions about your codebase in any channel. Passive watching captures decisions made in conversation.
@Kortex [project] [message] to capture, @Kortex [project] [question] to query - without leaving the conversation.
Works on every website. Gmail. Outlook. Notion. Word Online. When you mention a project, Kortex wakes up.
Completed tickets become context memories automatically. The intent behind every feature linked to the code that built it.
AI agents query your codebase intelligence before writing code. No more agents contradicting your team's decisions.
For individual developers
For growing product teams
For engineering organisations
Need unlimited scale? Talk to us about Enterprise →
Your competitors are shipping on guesswork. Their decisions disappear. Their new hires repeat old mistakes. Their AI agents contradict decisions made months ago. Yours do not have to.
Join the teams that never forgetSetup takes 5 minutes. The first intelligence captures itself.