Glasswarden · AI Governance, Security and Knowledge

Skip to content

GLASSWARDEN Set up a demo

Direct to Agent

Discover every AI tool and agent in your organization, enforce policy across every team, and remediate risk at its source.

Call 605-815-CRAF · Email sales@drycraf.com

Supported platforms and integrations

ChatGPT
Claude
Gemini
Microsoft Copilot
Claude Code
Codex CLI
Cursor
Gemini CLI
GitHub Copilot
OpenCode
MCP servers
OpenTelemetry
CrowdStrike Falcon
Defender for Endpoint
Microsoft Sentinel
Okta
GitHub
Splunk

Two sides to this coin

The frontier

America is leading the charge into AI. Now it's at your door.

AI operations are the biggest change in how work gets done since the internet, and bigger still. They're no longer only for the largest companies: small and medium-sized businesses can build, test and ship with coding AI assistants and agents working alongside their people today.

We don't fight vibe-coding. We shepherd it.

Build like a giant

AI assistants and agents let a small team work like a large one, and adopt new tools in days rather than quarters.

Claude Code, Codex CLI and Cursor on every laptop
New tools adopted in days
Small teams shipping like large ones

VibeCheck

Fixes you can push now: each one goes straight back to the agent that made the problem, Direct to Agent.

Propose, ask, or send with no approval
Closed only when telemetry shows the change
Make it policy, or accept the risk with an owner

VibeSec

What people build with AI, and how exposed it is, so speed never costs you your secrets or your customers' trust.

Every vibe-coded project, its runtimes and datastores
Findings by exposure, privilege and what's at stake
A risk score per project, person and team

Direct to Agent

Direct to Agent solutions

Glasswarden works where your people use AI and where your agents work. It sees what they do, controls what they're allowed to do, and fixes problems at the source, sending the fix back to the agent that made it.

01

AI Visibility

See every AI tool, agent, session and tool call.

The Glasswarden sensor runs where your people and agents work. It sees which AI assistants are in use, such as ChatGPT, Claude, Gemini and Microsoft Copilot, and agent hooks and an OpenTelemetry listener report what Claude Code, Codex CLI, Cursor, Gemini CLI, GitHub Copilot and OpenCode do. Glasswarden joins that with what your EDR, identity provider and code host already log.

An inventory of AI tools, agents, versions, models and MCP servers
Sessions and tool calls by person, team and device
Unsanctioned AI and shadow IT flagged as they appear

02

AI Control

The guardrails around what your agents do, written in plain words.

The Glasswarden sensor enforces your policies where AI and agents run. Each rule warns or blocks, for everyone or for one team, and an exception is granted to a person, team, agent or machine with a reason and an end date.

Natural language policy management: describe a rule, and Glasswarden writes it
Alerts from Glasswarden's rule packs
A sanctioned list of the AI services your teams may use
Policies learned from fixes, such as no public databases from agents

PoliciesNatural language policy management

AI read secret files into the chatRule pack: secretsAll teamsBlock

Hidden instructions in content the AI readRule pack: prompt injectionAll teamsWarn

New AI tool server addedRule pack: AI toolsEngineeringWarn

Storage bucket made publicRule pack: exposureAll teamsBlock Exception Data and analytics · “The data is a synthetic copy, not production” · until Dec 15 · granted by Daniel Reyes

Sanctioned AI services Claude Code 2.4+ · Codex CLI 0.50+ · Cursor 2.0+ · OpenCode 1.19+

03

AI Remediation

Direct to Agent: the fix goes back to the agent that made the problem.

Glasswarden writes the fix for the owner's own coding agent: the task, its limits, and the telemetry that will prove it done. The agent picks it up at its next session, reports progress back to Glasswarden, and the fix closes only when Glasswarden's telemetry shows the risk is gone.

Propose a change: the owner's reply is optional
Ask for change: the owner says yes, or no with a reason
No approval: a Fix admin sends it; applied at the owner's next session, and the owner is notified

High

Database anyone can readcustomer-portal · Maya Okafor · Claude Code on TRC-MAC-41A2

48 pts

The fix for Maya's agent

Close public access to the database

Turn on access rules for Supabase: no reads or writes without a signed-in user
Move the database key out of client code into a server-side setting
Rotate the key that was public

Queued
Delivered
Owner said yes
Agent working
Change pushed
Verified

Verified: Glasswarden's telemetry shows the risk is gone.

The Glasswarden dashboard

Watch it work.

A tour of the dashboard on sample data for Totally Real Corp, Inc.: its AI posture, alerts ranked by rule, a fix sent to an agent, telemetry and policies.

Glasswarden · Totally Real Corp, Inc. · sample data

Drive it yourself.

Open a lightweight copy of the dashboard with Totally Real Corp's sample data. Nothing to install, no sign-up.

Inside the platform

Every page, on sample data.

Convergence: The home page: ask Merlin anything about your organisation's AI use, or open Searches, Brief, Ions, VibeSec and VibeCheck. — ConvergenceThe home page: ask Merlin anything about your organisation's AI use, or open Searches, Brief, Ions, VibeSec and VibeCheck.

Ask Merlin: Merlin answers from what Glasswarden has confirmed, and cites each source. — Ask MerlinMerlin answers from what Glasswarden has confirmed, and cites each source.

The Glasswarden canvas: The organisation at a glance: risk, sessions, tool calls, alerts and unsanctioned installs, and the rules to focus on first. — The Glasswarden canvasThe organisation at a glance: risk, sessions, tool calls, alerts and unsanctioned installs, and the rules to focus on first.

Connections: Scroll out and the people, hosts and MCP servers behind each figure appear around it. — ConnectionsScroll out and the people, hosts and MCP servers behind each figure appear around it.

Overview: Scroll out further for the whole picture: teams, agents, silent sensors, rules firing and unapproved servers. — OverviewScroll out further for the whole picture: teams, agents, silent sensors, rules firing and unapproved servers.

Alerts: Every alert with its rule, agent and person, and whether a fix is ready to send. — AlertsEvery alert with its rule, agent and person, and whether a fix is ready to send.

Fix it with the agent: An alert opened: the detection, the calls around it, and a fix to send to the owner's coding agent at the approval level you choose. — Fix it with the agentAn alert opened: the detection, the calls around it, and a fix to send to the owner's coding agent at the approval level you choose.

Fix verified: The fix moves from queued to verified as the owner agrees, the agent works and telemetry confirms the risk is gone. — Fix verifiedThe fix moves from queued to verified as the owner agrees, the agent works and telemetry confirms the risk is gone.

Telemetry: Every agent tool call from every source, searchable by person, device, agent and tool. — TelemetryEvery agent tool call from every source, searchable by person, device, agent and tool.

Ledger: The facts your organisation knows, with who backs each one, when and from what. — LedgerThe facts your organisation knows, with who backs each one, when and from what.

Policies: Organisation-wide controls for sanctioned AI, MCP allowlists, secrets, prompt injection and tamper protection. — PoliciesOrganisation-wide controls for sanctioned AI, MCP allowlists, secrets, prompt injection and tamper protection.

Insights: How well the organisation's knowledge holds up, and how people use it. — InsightsHow well the organisation's knowledge holds up, and how people use it.

VibeSec: What people build with coding agents: where the code lives and runs, where the data goes, and what needs attention. — VibeSecWhat people build with coding agents: where the code lives and runs, where the data goes, and what needs attention.

VibeCheck: A queue of fixes for vibe-coded projects, ranked by risk, each ready to send to its owner's agent. — VibeCheckA queue of fixes for vibe-coded projects, ranked by risk, each ready to send to its owner's agent.

Ions: Long-running work across teams: what is blocked, what needs a second holder and who holds what. — IonsLong-running work across teams: what is blocked, what needs a second holder and who holds what.

Get started

Govern every AI agent from the first day.

Glasswarden brings AI governance, security and knowledge into one platform. Start with a demo on sample data, then connect your own AI and agents.

Set up a demo

Works with your AIChatGPT, Claude, Gemini and Microsoft Copilot; Claude Code, Codex CLI, Cursor, GitHub Copilot and more.
Fits your stackJoins your EDR, identity provider and code host, and forwards to your SIEM.
Built by DrycrafA company that defends information systems with cyber operations, security services and AI.