French Startup Edamame Builds Runtime Watch for AI Coding Agents
The platform uses host telemetry and AI analysis to flag intent drift, secret theft, and supply-chain interference in real time, before the damage lands.
Key points
- Edamame, a France-based startup, has launched a runtime verification platform for AI coding agents.
- It watches host-level telemetry for intent drift, secret theft, and supply-chain interference.
- Coding agents typically hold access to API keys and credentials sitting in environment variables or config files.
- No technical whitepaper has been published, so detection mechanics remain unverified.
- Jurisdiction falls under France's CNIL, with broader EU AI Act and GDPR obligations for enterprise customers.
What problem is Edamame solving?
AI coding agents write code, run commands, and pull dependencies with minimal human oversight. That autonomy is the point. It is also the problem.
Edamame is pitching a runtime verification platform designed to catch the moment an AI coding agent starts doing something it was never asked to do. "Intent drift" is the company's term for when an agent's actions diverge from its original task scope.
How does the platform work?
The platform watches at the host level, ingesting telemetry from the machine running the agent and feeding it through an AI analysis layer. Two of its three detection categories are relatively legible. Secret theft covers compromised or misbehaving agents quietly exfiltrating credentials. Supply-chain interference covers an agent fetching a malicious dependency or injecting code into a build pipeline, harder to catch after the fact and harder still to attribute.
Intent drift is subtler: behavior that isn't obviously malicious but isn't sanctioned either. An agent browsing outside its expected network scope, spawning unexpected child processes, or touching files unrelated to its task.
Edamame hasn't published a detailed technical whitepaper, so the precise mechanics of its detection logic, what telemetry signals it collects, how it baselines normal agent behavior, and what its false-positive rate looks like, remain unverified.
Should you worry about the gap this fills?
The product sits in a category that is forming fast. Our story on SymJack documented exactly how a malicious repository can turn an AI coding agent into a supply-chain weapon without triggering conventional endpoint alerts. Traditional detection tools were built for human operators. They weren't designed to distinguish "developer ran npm install" from "agent silently fetched a typosquatted package at 2 a.m."
As organisations pipe AI agents into CI/CD pipelines, runtime observability for non-human identities is a real gap. The judgment worth making plainly: Edamame is naming a genuine problem. Whether its detection logic is strong enough to match the pitch is something nobody outside the company can verify yet.
