← Back to Discover

Covin

Covin secures autonomous agents by checking what they do, starting with a free tool for coding agents on a laptop.

AI security & guardrails

What it does

Covin Local runs on a laptop. It checks each action that Claude Code, Codex or Cursor takes against the task the user gave it. Covin says Covin Local can pause or block actions that drift from that task. Every block comes with a reason. In the sample demo, a Claude Code session asked to change a button color has a read of ~/.aws/credentials blocked. Users can connect Covin Local to a free account at app.covin.dev to see what their agents did and anything that needs them. Covin Enterprise covers every agent in the company. It reads the records the cloud already keeps and joins them into a picture of who handed what to whom. Covin says it caught 91% of rule-breaking agents in simulations with scripted agents, five random seeds (range 87–96%) and no customer data. The same page reports 1.7% false alarms with policy checks and 7.6% with the statistical detector.

Capabilities

  • Covin Local lets the user pick a mode: observe, ask first, or block.
  • Covin says Covin Local blocks dangerous calls on the spot, even offline.
  • Covin Local has a rules feature where users type in plain words what their agents should never do. Covin says it drafts the rule and shows it to the user first.
  • Users can ask Covin why it made a decision and get one decision explained in plain words, with --details for more.
  • Covin Teams, in beta, gives every teammate's laptop a shared dashboard at teams.covin.dev with the roles admin, viewer (read-only) or laptop only.

When to use it

  • You want to stop a coding agent that was asked to fix CSS from reading your cloud keys.
  • A startup or small team wants every teammate's laptop to report to one shared team workspace.
  • A company wants to trace agent workflows from a developer's laptop through CI to its cloud accounts.