Frequently asked questions
Short answers to the questions security teams and developers ask most.
Data and privacy#
Does Vardhold send anything to Dawnstar Labs or a cloud service?
No. Vardhold runs entirely on the endpoint and needs no account or outbound connection. Events leave the machine only for the SIEM destinations you configure.
Does Vardhold read my prompts or my code?
It sees the actions an agent takes: the commands it runs, the paths it reads and writes, the hosts it contacts and the MCP tools it calls with their arguments. It does not record the conversation with the agent or the contents of files. Secrets in commands and arguments are redacted before anything is stored.
Where is the data kept, and for how long?
In %ProgramData%\Vardhold, readable only by SYSTEM and Administrators, for 30 days by default. Your SIEM keeps the long-term copy.
Developers#
Will it slow my agent down?
A Claude Code check adds about 45 ms. Most actions are logged without interruption; only the few your policy marks ask wait for a person.
Can a developer, or an agent, turn Vardhold off?
Not without an administrator. Lowering the mode, releasing the kill switch, changing the policy or settings, and removing the hooks all need administrator confirmation, and the default policy blocks agents from stopping or removing Vardhold. Making things stricter is open to anyone.
What does the agent see when it is blocked?
A clear refusal naming the rule and the reason, so the agent can explain it to you and choose another way.
Coverage#
Which agents does Vardhold support?
Claude Code, Claude Desktop, Cursor, Windsurf, VS Code with GitHub Copilot, GitHub Copilot CLI, OpenAI Codex CLI, Gemini CLI, Aider, Ollama, LM Studio, OpenClaw and any MCP server. Claude Code, Cursor and MCP servers are checked before each action; the others are recorded as they act. See Connect your agents.
Does it work on macOS or Linux?
Not yet. Vardhold is built for Windows, where most enterprise and government endpoints are. Tell sales@dawnstarlabs.com if you need another platform.
Does it replace our EDR?
No, it complements it. EDR sees processes and malware. Vardhold knows which AI agent acted, checks the action against your AI policy before it runs, involves a person when it matters, and records the decision. Both feed the same SIEM.
Buying and compliance#
Is Vardhold certified?
No. It is designed to support controls in NIST SP 800-53 and other frameworks, and ships with a hardening checklist, a cryptography statement and an SBOM for your package. See Compliance alignment.
How do we try it?
Through a 60-day pilot: we help you install it on a group of developer machines, tune the policy with you, and review the results together. Start a pilot.
What does it cost?
Pricing for pilots and deployments is agreed with each customer. Email sales@dawnstarlabs.com for a quote.
What is coming next?
Pre-execution coverage for more agents, central management through Group Policy, Intune and an on-premises console, a home edition, and kernel-level prevention for agents without hooks.