Corteza – Capture team decisions and makes them searchable with AI

Name: Corteza – Capture team decisions and makes them searchable with AI
Availability: InStock
Author: ctumani

by ctumani·Feb 25, 2026·2 points·0 comments

Visit Project View on HN

AI Analysis

●MidCrowd PleaserSolve My Problem

Decision bot for Slack threads, but Notion, Confluence, and internal wikis already solve this cheaper.

Strengths

•Captures decisions at point-of-entry (Slack) minimizes friction vs. requiring external tool
•Semantic search via vector embeddings finds intent even with different wording
•Early-stage but ships with demo and 2-minute onboarding, real Slack/Claude 3.5 integration

Weaknesses

•No GitHub/email integration—Slack-only limits utility for non-Slack-first teams
•Doesn't differentiate from wiki + search; paying for MongoDB Atlas + Claude calls when Notion or Confluence + OpenSearch are cheaper per seat

Post Description

Hey HN, I'm Cristian, founder of Corteza (https://corteza.app).

The problem: product teams lose decisions in Slack threads, Miro boards, Confluence pages, meetings, etc. Three months later nobody remembers why something was built a certain way, and the team re-litigates the same calls over and over.

What Corteza does: you type /decision [what] [why] [alternatives considered] in any Slack channel, or via our Chrome Extension. The bot stores it with full context. Later, anyone can ask natural language questions like "why did we pick PostgreSQL?" and get back the exact decision with reasoning, who made it, and when — even if they use completely different words.

Tech: Node.js + Slack Bolt SDK, MongoDB Atlas Vector Search for semantic retrieval, Claude 3.5 Sonnet for the conversational interface, Railway for hosting.

There's a live demo (no Slack required): https://app.corteza.app/demo

Early stage — looking for product teams who are frustrated by this problem and want to try it. Would love feedback from the HN crowd especially on the retrieval approach (we use OpenAI embeddings + cosine similarity, then Claude re-ranks and explains).