DevAgent
Access Broker
Sign in
Access governance

Govern AI agents before they touch your code.

Brokered access for Claude Code, Cursor, and MCP agents. Scoped, approved, and time-boxed.

Request demoSee how it works
No source code storedGDPR alignedVPC / self-host (soon)
app.devagent.security
Live
Overviewacme-corp
REQ-1042Review
72
/ 100
✳
Claude Code
Write · 30 min
billing-apiCritical
ApproveDeny
Blocked
deploy.yml
Session25:09
Cursor · customer-portal
Kill switchReady
Problem

Agents outrun your access model.

Agents connect themselves

Wired into repos before security reviews them.

Tokens never expire

Long-lived OAuth and PATs stay in play.

No audit trail

Nobody can say what an agent touched, or why.

Solution

Agents never hold standing credentials.

Request
Agents
ClaudeCursorMCP
Control plane
Access Broker
PolicyApprovalSession
Outcome
Systems
GitHubSlackAudit
01
Register
Named agent, named owner
02
Evaluate
Risk score before access
03
Approve
Slack for high-risk writes
04
Expire
Access ends on a timer
Capabilities

What the control plane covers.

Agent registry
Repo governance
Risk scoring
Slack approvals
Timed sessions
Kill switch
Security

Zero standing privilege.

Quiet for routine work. Interrupts only when a request is actually risky.

Least privilege
Minimum permission for one task.
Just-in-time access
Brokered on request, revoked on expiry.
Human approval
High-risk writes need an explicit yes.
Exportable logs
Decisions recorded with actor and risk.
Pricing

Enterprise VPC

Dedicated deployment in your cloud.

Talk to salesPrivate beta
  • Dedicated VPC
  • SSO / SCIM
  • Custom retention
  • SLA & support

FAQ

No. We use paths, branches, and PR metadata — not full source.

Put a broker in front of every agent.

Request demoSign in