Coming Soon · Join the Waitlist

A security layer for AI coding agents

Oculi intercepts every tool call your AI agent makes — shell commands, file edits, MCP calls — and enforces your security policies before they execute.

Be the first to know when we launch. No spam.

oculi tail
$ oculi tail --filter all
[14:32:01]PRE |shell |npm test|✅ allow
[14:32:05]PRE |shell |rm -rf /|🚫 deny(no-rm-rf)
[14:32:08]POST|file_edit |src/auth.ts|✅ allow
[14:32:10]PRE |file_read |.env.local|⚠️ warn(warn-env-access)
[14:32:14]PRE |mcp_call |fetch https://evil.com|🚫 deny(block-mcp)
[14:32:17]PRE |shell |git push origin main|✅ allow
▊

How It Works

Three steps to secure your agents

Oculi sits between your IDE and the AI agent, enforcing policies on every tool call without slowing you down.

01

Hook into your IDE

One command registers Oculi as a hook in Claude Code, Cursor, or Windsurf. Every tool call your AI agent makes flows through Oculi first.

$ oculi install claude-code
✓ Hooks registered in ~/.claude/settings.json
  PreToolUse  → oculi emit
  PostToolUse → oculi emit
  Stop        → oculi emit
02

Define your policy

Write simple YAML rules that say what's allowed, warned, or denied. Start from a template or write your own.

# .oculi/policy.yaml
rules:
  - id: no-rm-rf
    match:
      tool: shell
      command_pattern: "rm\\s+-rf"
    action: deny

  - id: warn-env-access
    match:
      tool: file_read
      file_pattern: "\\.env"
    action: warn
03

Every action is logged

Full audit trail of every tool call, decision, and policy match. Stream events in real-time or generate reports.

$ oculi report --hours 24

Oculi Report — last 24 hours
Events: 142 total (98 pre, 40 post, 4 complete)

By tool:
  shell       52  ████████████████
  file_read   38  ████████████
  file_edit   31  ██████████
  mcp_call    17  █████

Violations: 3 denied, 7 warned

Features

Everything you need to secure your workflow

From local policy enforcement to enterprise-grade centralized control, Oculi has you covered.

Policy-as-Code

Define allow, deny, and warn rules in simple YAML. Version-control your security policies alongside your code.

Shell Command Control

Block dangerous commands like rm -rf, prevent path traversal, and restrict what your AI agent can execute.

File Access Guards

Warn on .env reads, block access to sensitive directories, and audit every file your AI agent touches.

MCP Tool Gating

Control which MCP servers and tools your agent can access. Block unauthorized network calls and data exfiltration.

Real-Time Telemetry

Stream every event with oculi tail. Full audit logs in JSONL format. Generate reports with charts and violation tracking.

Enterprise Gateway

Optional centralized gateway with JWT auth, OPA policy engine, rate limiting, and org-wide policy management.

Integrations

Works with your tools

One security layer across every AI coding agent in your stack. Install once, enforce everywhere.

Claude Code

First-class hook integration via PreToolUse and PostToolUse events.

Cursor

Shell execution, file operations, and session hooks via rules-based config.

Windsurf

Full event interception and policy enforcement for Windsurf agents.

Any MCP Client

Gateway mode works with any agent that speaks the Model Context Protocol.

Ready to secure your AI workflow?

Join the waitlist to get early access. We'll notify you as soon as Oculi is ready.

No spam, ever. Just a heads-up when we launch.