Karim Rahme on Metorial's open-source enterprise integration layer for AI agents: 3,600 GitHub stars in 5 weeks
Dec 3, 2025 with Karim Rahme
Key Points
- Metorial, a YC-backed open-source integration layer for AI agents, hit 3,600 GitHub stars and 1,000 weekly active users within five weeks of launch, with seed funding closing in five days.
- The startup's core advantage is model neutrality and granular access control, letting enterprises run multiple LLMs against apps like Salesforce and Gmail without exposing all data to every employee.
- Rahme positions Metorial as infrastructure substrate for any AI agent needing to read and write enterprise data, betting the integration layer itself matters more than which protocol becomes standard.
Summary
Karim Rahme is the founder of Metorial, a YC company building an open-source integration layer that gives AI agents access to enterprise apps and data sources — think Gmail, Salesforce, SAP — with granular access control baked in. Five weeks after launch, the project has 3,600 GitHub stars and close to 1,000 weekly active users. The company is in final-stage discussions with Fortune 500s and unicorns planning to deploy it across organizations of 80,000 to 100,000 members. The seed round closed in roughly five days.
The core pitch
The wedge against OpenAI and other LLM providers building their own integrations is model neutrality. OpenAI won't build integrations for Gemini or Anthropic, so enterprises that want to run multiple models need a provider-agnostic layer. Metorial's other differentiator is access control — Fortune 500s can't let an LLM touch every Salesforce record for every employee, and Metorial handles the permissioning logic that generic integrations don't.
MCP positioning
Rahme's view is that companies betting entirely on MCP are following hype rather than solving the underlying enterprise need. Metorial uses MCP as middleware today, but the architecture is designed to swap in whatever protocol becomes standard. The long-term bet is on the integration and access-control layer itself, not on any specific protocol.
The longer ambition is an Oracle-style substrate story — own the integration and access-control foundation, then layer on workflow builders and agent hosting on top.
Rahme's read on 2026
He's most excited about what he calls full-stack AI-native firms — small teams using legal or healthcare agents to compete directly with established players and unicorns. That framing also maps to Metorial's commercial thesis: every agent, regardless of vertical or architecture, needs to read and write to apps and data sources, and Metorial wants a cut of that.
Team
Rahme and his co-founder met at an Austrian technical high school at 14 and have over 11 years of formal computer science education between them. They've built Metorial's infrastructure from scratch rather than relying on off-the-shelf tooling — a deliberate choice Rahme argues gives them a technical edge over competitors built primarily on vibe coding.