Agentic Runtime

Agentic Runtime protects the AI coding agents running on your developers' machines, such as Claude Code, Codex, Cursor, Aider or Gemini. A lightweight agent on each device inspects what those coding agents send and do, and enforces the policies you define in the console.

Limited access

Agentic Runtime is available to select customers. Contact sales to request access.

What It Catches

Family	What it protects against
Credential leaks	AWS, GitHub, Anthropic, Stripe and Slack tokens, JWTs, PEM keys and 100+ more, redacted before they leave the device
Prompt injection	Jailbreaks, system-tag spoofing and hidden instructions in loaded context: READMEs, pasted documents, tool results, MCP responses
Dangerous tool use	`rm -rf`, `curl
Malicious MCP servers	Known-bad servers, typosquats, vulnerable versions and scope abuse, caught before they connect
Risky skills	Compromised `SKILL.md` files, shell-exfiltration patterns, untrusted authors and dangerous tool grants
LLM models	Models or entire AI providers your organization does not allow
Sensitive repository paths	Read, edit, write and search calls against paths you deny

Prompts never leave the device. Only redacted incidence metadata reaches the console.

How It Works

Each developer installs the agent and links the device to your organization.
The agent wraps the coding agents on the machine through shell aliases, with no environment variables or flags to set.
It inspects prompts, loaded context and tool calls on the device, and applies the effective policy for that device.
Every 30 minutes it reports the device's inventory: MCP servers, skills and the repositories it touched.
Detections appear in Incidences, and every policy change in the Audit log.

Policies That Follow Your Organization

Policies cascade from the broadest scope to the most specific:

TEXT

Organization  →  Team  →  Member  →  Device

Organization  →  Team  →  Member  →  Device

Each level can override the one above it. When a member belongs to several teams, the strictest policy wins on each axis. See Policies.

The Console

Agentic Runtime lives under Agentic runtime in the ByteHide console:

Group	Screens
—	Incidences
Policies	Organization, Teams, Users, Devices, Repo, Custom Rules
Inventory	MCPs, Skills, Tools, Models (Agents coming soon)
Administration	Audit log, Settings

Configure Your Policy by Voice

Every screen offers Configure your policy by voice. Tell the assistant what you want to detect and block, and it prepares a policy proposal for you to review and apply. Nothing changes automatically.

Agentic Runtime and AI Runtime

	Agentic Runtime	AI Runtime
Protects	The AI coding agents your developers use	The AI applications and agents you build and run
Runs	On each developer's device	As a gateway, or in-app through the App runtime SDK
Scope	Organization, team, member, device	Organization, per route