Operational memory for AI coding agents

Give Claude Code your team's memory.

Cortex turns your repo, agent sessions, docs, and past mistakes into living operational context — injected into Claude Code before it acts.

1,843 engineers already in line
memory core · online
cortex · session
> claude code session
  ↳ Cortex learns
  ↳ memory approved
  ↳ future agents get smarter ✦

Every session starts from zero

Without Cortex

Agents forget how your company works.

They edit files and ship changes — but never learn your architecture, deploy rules, or past outages. So they repeat the same mistakes and need constant babysitting.

With Cortex

Every session makes the next one smarter.

Cortex captures what works from real engineering, turns it into structured memory, and injects the right context into Claude Code exactly when it acts.

Everything your agents should already know.

◈

Repo Memory

Project conventions, architecture, commands, risky files, and deployment rules — structured and queryable.

⟁

Session Learning

Capture what Claude Code tried, what failed, what worked, and what should be remembered.

❖

Memory Inbox

Review, approve, reject, edit, and scope every new memory before it reaches an agent.

⚠

Risk Warnings

Warn agents before they touch sensitive files or repeat a known outage.

✦

Self-improving reviews

The memory-grounded PR reviewer learns from every accept or dismiss — the memory behind each finding gains or loses trust, so reviews sharpen over time.

❒

Living Docs

Generate onboarding guides, runbooks, and architecture notes from actual work.

⌁

MCP Integration

Expose memory directly to Claude Code through the Model Context Protocol.

The learning loop

Cortex doesn't just remember — it learns from your reviews.

Every finding from the memory-grounded PR reviewer is anchored to the memory that justified it. When you accept a finding, that memory earns trust. When you dismiss one, it loses trust. So the reviewer stops repeating noise and gets sharper with every pull request.

  • ✓Accepted — the grounding memory gains confidence and surfaces sooner next time.
  • ✕Dismissed — its confidence drops, so the noise fades away.
  • ✦A memory fired — reviews compound, tuned to how your team actually works.
cortex review · acme/billing-api#482
▸ finding · idempotency key missing on refund path
  ↳ grounded in: "Duplicate invoice outage — March"
  ✓ accepted by @maya
  ✦ memory confidence 0.62 → 0.81

▸ finding · rename shadows exported symbol
  ✕ dismissed (intentional)
  ✦ memory confidence 0.48 → 0.30

reviews get sharper every PR ✦

How it works

  1. 1

    Install Cortex in your repo

  2. 2

    Connect Claude Code through MCP and hooks

  3. 3

    Let agents work normally

  4. 4

    Review proposed memories

  5. 5

    Future sessions become smarter

Get started

Connected to Claude Code in under a minute.

Install the CLI, sign in, and link your repo. Cortex registers an MCP server so Claude Code can retrieve your team's memory automatically.

  • ✓ Works with your existing repos
  • ✓ No code changes — just `.mcp.json` + `CLAUDE.md`
  • ✓ Memory stays scoped to each repo
Full setup guide →
your-repo — setup
# 1 · install the CLI
npm install -g @mxbenjaminbeguin/cortex

# 2 · authenticate
cortex login

# 3 · connect this repo
cortex init

✓ wrote .mcp.json, CLAUDE.md, hooks
✓ Claude Code now has your repo's memory
search_memoryget_repo_contextrecord_session_summary

Before it touches code

Claude knows the history before editing billing.

Memory is retrieved and scoped to the repo and task — so the agent walks in already knowing the landmines.

get_repo_context · acme/billing-api
Relevant context:
- Billing uses Stripe webhooks.
- Duplicate invoice outage happened in March.
- Always check idempotency keys.
- Run make test-billing before opening a PR.
- Do not edit invoices_v1 tables.

Built for AI-native engineering teams.

  • ▹Startups using Claude Code
  • ▹Teams with fast-moving repos
  • ▹Engineering orgs with tribal knowledge
  • ▹Platform teams managing risky workflows
  • ▹Founders who want agents to work safely

Make every AI coding session compound.

Join the waitlist and be first to give your agents a memory that grows with your team.

1,843 engineers already in line