The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Back to feedPost
Engram — Self-Hosted Company Knowledge Brain
What it does
Engram is an MCP (Model Context Protocol) server that turns your company's scattered knowledge into a unified, queryable brain. It ingests data from Slack, Notion, Gmail, and GitHub — and makes all of it instantly accessible to AI assistants like Claude.
The Problem it Solves
Engineers waste hours hunting through Slack threads, Notion pages, and old PRs just to answer "how did we decide this?" Engram fixes that. Ask once, get the answer instantly — with full source attribution.
Technical Architecture
Vector Store: PostgreSQL + pgvector for semantic search
Graph Layer: Kuzu graph database for relationship mapping between documents and people
MCP Server: Native integration with Claude and other AI assistants
Ingestion Pipelines: Slack, Notion, Gmail, GitHub connectors
Self-hosted: Full data privacy, runs on your own infrastructure
Key Features
🔍 Semantic search across all connected sources
🕸️ Graph-based relationship mapping (who said what, linked to which doc)
🔒 Fully self-hosted — your data never leaves your infra
⚡ Real-time ingestion from live workspace tools
🤖 Plug-and-play with Claude via MCP protocol
Built For
Startups and engineering teams that want AI that actually knows their codebase, decisions, and context — not generic answers.
#RAG #LLM #AIEngineering #EnterpriseAI #MCP
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Related posts
Wanted to say a bit more about our first template and share a glimpse into our process.
If you are interested in how to build framer templates feel free to read and ask questions in the comments.
really nice work
Title:
I built a design system generator inside Figma Make. Here's the workflow.
The problem I kept running into: every new client project meant re-explaining the brand to every AI tool from scratch. Figma Make didn't know the colors. Cursor didn't know the tokens. Claude didn't know the rules. I was the shared memory — and I kept losing things.
So I built Design System Gen. Entirely inside Figma Make. Powered by Claude API.
The workflow:
Upload brand references → Claude API analyzes them across 8 dimensions (palette, mood, typography, composition, lighting, texture, motion, UI DNA) → structured data, not vibes → generates 11 files, each formatted for a specific tool in your stack.
figma-make.md → paste once, Figma Make knows your entire design system
design-system.css → all tokens as CSS custom properties
tokens.json → W3C format, imports directly into Figma Variables
starter.tsx → React starter with tokens already bound
.cursorrules → Cursor knows your system from root
Claude.md → Claude reads it automatically every session
+ 5 more for docs, LLMs, prompts, components
Every tool gets its own context file. All consistent with each other.
The part I'm most proud of: I'm a designer. I don't write code. The entire tool — UI, API integration, file generation logic — was built and shipped inside Figma Make. No external hosting. No separate codebase.
Live at: vibecodingdsgen.figma.site
Add your Claude API key and try it with your own moodboard.
#FigmaMake #DesignSystem #VibeCoding #AI #Claude #UXUI #DesignSystems #ConfigMakeathon
As someone who constantly jumps between Figma, Claude, ChatGPT, Cursor, Lovable, and other AI tools, this problem feels very familiar 😅
The tools are getting smarter, but keeping context consistent is still surprisingly manual. Love seeing people tackle that gap.
Figma Make isn't just a prototyping tool. I used it to ship a Claude-powered design system generator.
Most designers treat Figma Make as a quick vibe-coding sketchpad. I wanted to show it's something deeper — a full vibe-coding platform that integrates with external APIs.
The connection nobody writes about
There are three ways Claude connects to Figma today. Figma MCP — Claude reads and writes Figma files via MCP server. Figma's built-in AI — Figma calls its own models internally. And mine — a React app inside Figma Make that calls Claude API directly from the browser. Claude responds to the code, not to Figma. Figma Make is the hosting.
The new pipeline
Old approach: get brandbook and references from client → generate moodboard → align → write prompts for UI Kit. Separate steps, human in the center every time.
My approach: collapse all of that into one step. Run an AI analysis and output structured data — input for machines, not a result for humans.
We don't ask AI how the brand looks. We run an MRI on it and get structured data that machines use to build code.
What gets generated
11 files, all consistent with each other because they share one analysis. For Figma specifically:
tokens.json — W3C DTCG 3-tier format (primitive → semantic → component), imports directly into Figma Variables via Tokens Studio or Figma Make prompt
design-system.css — all tokens as CSS custom properties, ready to import
starter.tsx — paste into Figma Make, full design system renders instantly
figma-make.md — session context, paste once and the AI knows the entire design system
Add a content plan on top — generate a full design in one run with minimal token spend.
The entire tool was built and shipped inside Figma Make. No external hosting. No separate codebase.
#ConfigMakeathon #FigmaMake #DesignSystem #VibeCoding #ClaudeAPI #BuildInPublic
tNot gonna lie, I opened this out of curiosity and left with more questions than answers 😅 But that's usually a good sign. Really interesting exploration of what design workflows might look like in the AI era. And by the way, this link doesn't work figma-make.md
Trending
Claude
Claude has entered the design space. How are you using Claude Design?
Contra University
Learn from expert creatives how to earn more using next-gen AI tools.
MagicPath
The canvas is infinite, and exploration is becoming the workflow. How are you using MagicPath?
creativeaiflow
Creative AI workflows are evolving. What tools do you use, and what are their strengths and weaknesses?
freelancerlife
Freelancer life is wins, pivots, and everything in between. What’s yours right now?
