Stop AI coding agents before they make unsafe changes.

On supported workflows, Dexgate evaluates covered shell, Git, and deployment actions before execution. When policy permits, it issues a scoped Action Passport and preserves reviewable evidence.

See current coverage →

See the interactive Action Passport demo →

npx --yes --package=@dexgate/openclaw-trusted-mode@1.0.18 dexgate-assess

Checks the packaged policy and looks for an existing OpenClaw configuration. It does not install the adapter or prove that a live agent action was intercepted. What this means →

Current recommended version: 1.0.18

Current availability

OpenClaw Recommended

Free adapter available now

Governed workflows available through the early-access pilot

Free steps Demo

Codex Free available

Within documented coverage

Governed workflows available through the early-access pilot

Free steps Demo

Free adapters: available now on npm

Paid governed workflows: early-access pilot

Coverage last validated: July 30, 2026

Validated free packages: @dexgate/openclaw-trusted-mode@1.0.18, @dexgate/codex-trusted-mode@0.1.16

Roadmap: Grok Build, Claude Code, Cursor, GitHub Copilot.

See the compatibility matrix for coverage, limitations, and current versions.

Free → governed

Two levels of control. One plugin.

Install once. Choose how decisions are made. Free mode is local and offline. Governed mode adds real policy evaluation, evidence, and fail-closed authorization when you are ready.

Free Governed
Best for Day-one safety, demos, personal agents Teams, pilots, production controls
Decision path Local allowlist on the machine OpenClaw → Decision Middleware → Dexgate runtime
Network None Your decision services (local or private)
What gets blocked Tools not on your list Policy-based allow / deny / constrain
Evidence — trace_id + reconstructible evidence per call
If decision path is down N/A (local only) Fail-closed (block by default)
Default? Yes Opt-in config

Free hard gate

Safer OpenClaw defaults without accounts or outbound calls. High-risk tools stay off until you expand the list.

Governed path

Real policy evaluation with reasons you can log and review. Decision Middleware returns decisions and evidence (passport: null on /v1/decide today). Entitled licensed-runtime workflows can issue and verify Action Passports separately.

Ready for real decisions?

Keep @dexgate/openclaw-trusted-mode@1.0.18. Run @dexgate/decision-middleware@0.1.1 and your licensed Dexgate runtime. Set toolPolicyMode to MIDDLEWARE—no adapter swap.

  1. Keep the free plugin installed (1.0.18+)
  2. Run Decision Middleware + licensed Dexgate runtime
  3. Set toolPolicyMode: "MIDDLEWARE" and your decide URL

See the governed path Start Production — $99/month Compare all plans

The public npm plugin is MIT. The Dexgate runtime is licensed separately. Free mode never requires the licensed runtime.

Interactive demo

See a policy decision and Action Passport flow

Illustrative walkthrough for supported surfaces. Expand the panel to step through the flow.

Interactive demo

See a policy decision and Action Passport flow in 30 seconds

Illustrative example — not a captured customer event. This walkthrough does not call a live Dexgate runtime.

Proposal
Codex proposes apply_patch("settings.py") against production configuration.
Policy Evaluated
Production-change profile: evidence, environment, target, and rollback constraints.
Policy decision
DENY: missing rollout evidence
Authorization / execution verification / outcome record
Authorization: none; execution verification: refuse; outcome: executor did not mutate files.

On the paid pilot path, each governed decision should leave proposal, policy decision, Action Passport or no-passport refusal, execution verification, and outcome evidence.

Without passport governance

Agent inherits permissions

Action executes

Monitoring sees it later

With governance

Action is intercepted

Policy returns a policy decision

Executor verifies or refuses a scoped passport

Agent Proposal

> apply_patch("settings.py") in production

Illustrative example — not a captured customer event. The paid path is useful only when the protected executor can verify a scoped passport or refuse the action.

Choose your adapter

Pick the adapter for the runtime your team uses today. Start with a free path on npm for evaluation, then use the paid pilot when you need Passport evidence.

OpenClaw adapter

OpenClaw — recommended free entry path

Recommended free entry path on npm. Optional policy self-check in about a minute. A complete first adapter setup typically takes 10–15 minutes after prerequisites are installed.

Codex adapter

Codex — available free within documented coverage

Conservative read-only shell hard gate; patching and mutating commands blocked by default. Controlled host coverage.

Roadmap: Grok Build, Claude Code, Cursor, and GitHub Copilot — same free hard-gate model when published.

See all adapters · Run the free policy self-check

Optional free policy self-check (from the compatibility matrix):

npx --yes --package=@dexgate/openclaw-trusted-mode@1.0.18 dexgate-assess

PASS means the packaged free policy handled representative actions as expected—not that OpenClaw enforcement is live. Install the free OpenClaw adapter for real sessions. See the governed path when you need shared policy and reviewable evidence.

Why the control layer exists

Traditional security controls who can access systems. AI agents inherit those permissions and can act autonomously within them—so access control alone is not enough.

Sandboxing changes where an agent runs. It does not decide whether a sensitive allowed action should proceed.

Adjacent approaches leave a gap at the action boundary:

Approach Limitation
Permissions The agent still inherits the user's or service account's allowed actions.
Sandboxing It limits environment reach but does not approve or deny business-sensitive actions.
Agent settings They shape behavior, but they do not provide independent action approval or evidence.
Monitoring It helps with forensics, but usually only after the action has already happened.
Policy engines They express and evaluate policy, but still need agent-path integration, enforcement at the action boundary, and a surrounding evidence model.

On supported integrations and configured surfaces, dexgate evaluates covered agent actions before execution and leaves a reviewable record behind.

Decision Execution Governance

Decision Execution Governance is the missing control layer for organizations running automated systems and AI agents. Instead of only controlling access or monitoring after the fact, for supported integrations and configured execution surfaces it evaluates covered actions before execution and returns an allow, constrain, deny, or escalation result—with a reviewable decision record for engineering, security, compliance, and audit teams. Current adapter and surface coverage is defined on the compatibility matrix.

Traditional Security

User
AI Agent
System

Controls access

Decision Execution Governance

User
AI Agent
Governance Layer
System

Controls decisions before they become actions

Agent Execution Governance

Agent Execution Governance applies Decision Execution Governance specifically to AI agents.

It gives security and engineering teams action-level control and visibility beyond the permissions inherited from human users, for autonomous agents on supported integrations and configured surfaces.

Traditional security controls who can access systems. Decision Execution Governance controls what automated systems and AI agents are allowed to do once they have that access.

From local to governed

How Dexgate works

Same OpenClaw plugin either way. Free mode is the default hard gate; governed mode is a config change when you are ready.

1

Intercept the action

On supported hosts, the adapter sees the proposed tool call before the runtime commits to it.

2

Free hard gate (default)

ALLOWLIST_ONLY blocks tools not on your list—offline, no account, no network.

3

Governed evaluate (opt-in)

MIDDLEWARE mode: Decision Middleware → Dexgate runtime for allow / deny / constrain with environment-aware policy.

4

Leave evidence behind

Governed outcomes include trace_id and reconstructible evidence. Fail-closed if the path is down. Action Passports apply on entitled paid workflows.

Install the free OpenClaw adapter See the governed path Compare all plans

Evidence and assurance

Evidence for engineering, security, and audit review

Give security, compliance, and audit teams structured evidence they can use during rollout review.

Two kinds of artifacts: records the product can generate on the paid governed path, and public templates or first-party review materials. They are not the same. These materials support internal review—not regulatory compliance or third-party approval by themselves.

Product-generated evidence

On the entitled paid path for supported surfaces, governed runs can produce customer-specific records:

Generated by the product

Policy decision records

Generated by the product

Action Passport records

Generated by the product

Authorization-verification records

Generated by the product

Execution or refusal outcomes

Generated by the product

Governed-action history

Review materials and templates

Public downloads help you prepare a review. They are not automatic per-customer runtime outputs.

View Assurance Details Open Review Kit

Commercial Model

Pricing

Start free with local hard gates. Move into dexgate when you need governed action control, decision records, and managed policy updates.

One pricing ladder for supported paid integrations (OpenClaw recommended free path; Codex available free). Coverage and plan details: Compatibility, Pricing, and Assurance.

Free

$0

Free local hard gate for supported adapters.

Production

$99/mo or $990/yr

1 protected deployment / 2 environments.

Team

$499/mo or $4,990/yr

3 protected deployments / 3 environments.

Business

$1,999/mo or $19,990/yr

10 protected deployments / 5 environments.

Enterprise: custom annual agreement for procurement, rollout design, and expanded validation scope.

Start Production — $99/month Compare all plans Guided pilot — $3,500

Roadmap

OpenClaw and Codex free adapters are on npm today. Additional agent runtimes are on the roadmap. Next: broader adapter coverage, CI governance, and richer evidence packaging for security review.

CI Governance

Deterministic controls for promotion, release approval, and deployment boundary changes.

Additional Runtimes

More governed adapters for agent and automation surfaces beyond today’s coding runtimes.

Evidence Automation

Cleaner packaging for review kits, release notes, validation history, and security review packs.

Get Updates

Subscribe for category updates, release notes, and early adopter news without booking a sales call.

Subscribe

Enterprise Support

Self-Serve First, Enterprise Ready

Install and verify quickly, then bring in partner-assisted rollout planning, buyer review support, and security evaluation guidance when your organization needs it.

Rollout design Buyer review support Security evaluation guidance

Start a guided pilot View implementation partners