Built for the way you actually think
Cortex is a founder-built AI Second Brain. It exists because no existing tool let me converse with the knowledge I’d already collected — and because I wanted an agent loop I could actually own.
Why this exists
I’d been collecting YouTube videos, articles, transcripts, and voice notes for years. Across Notion, Mem, Obsidian, three different read-later apps, four different note-taking apps. The collection grew. The understanding didn’t. Search returned files; it never returned answers.
Then the agent-SDK era hit. Suddenly the missing piece — a model that could call tools, read its own context, and reason across sources — was a pip install away. I started prototyping in early 2026. The first version of Cortex was a single FastAPI file plus a Whisper wrapper. It surprised me with how well it worked.
The thing that kept surprising me: an MCP-native architecture with a CodeAct orchestrator, a 4-layer memory (full-text + vector + claims + nightly REM consolidation), and real entity extraction is fundamentally different from a full-text search box bolted onto a notes app. The conversations go deeper. The connections are real. You can ask “what did I learn last quarter that contradicts what I thought yesterday?” and get a citation-backed answer.
“I built Cortex because I wanted to talk to my knowledge — not search it. The fact that the agent loop is fully owned, BYOK-able, and runs on my own infra was the price of admission, not the feature.”
— Manik Rai, founder
Principles
- Own your loop. BYOK Anthropic / Gemini / OpenAI keys, your own MCP tools, your own data on your own SQLite or Turso. No vendor lock.
- Less is more. Use the best library for each job (WorkOS, Composio, Stripe, shadcn) instead of reinventing. Keep custom code minimal and honest.
- Citations, always. Every answer the agent gives is backed by the chunks + entities + memory facts it pulled. No hallucinated confidence.
- Built in public. Open Jira board, public sprint logs, weekly OWASP MCP self-assessment. The work is the marketing.
What’s next
Phase 7 — Cortex Teams — is shipping in pieces through Q2 2026. Foundation, CRUD, billing (Stripe Pro Team @ $18/seat), and team-scoped memory are landing in sequence. After that: Cognee deep-KG mounting, V-JEPA dense embeddings (research), and a paid pen-test before public launch.
If any of this resonates, get on the early-access list. We’re onboarding manually right now — every signup gets a real reply from me.