See consequential actions
Know what an agent is about to do before it does it. Runtime captures tool calls, shell commands, file access, network destinations, and sensitive data movement at the control point.
While agents act
Control risky agent actions. Before they execute.
Runtime is the part of Signals Corps that checks AI agent actions while they run. Recruiters, sales teams, and engineers keep working, and uploads of candidate files, customer data leaks, and risky commands stop before they happen. Runtime looks at tool calls, file access and uploads, shell commands, and network activity in the flow of work, then allows, warns, asks for approval, or blocks.
Free devices on every plan, then priced per device. See pricing
Runtime
Run the setup steps from the README
Bash curl -s https://get.example.dev/setup.sh | bash
Blocked by Signals Corps Runtime
- Rule
- Remote script piped to a shell
- Severity
- High
- Decision
- Blocked before it ran
- Reported to
- IT & Security teams
I didn't run it. Signals Corps blocks piping a remote script into the shell. I can download it so we can read it first.
Decisions in the moment
Security that understands prompts and the actions they trigger.
A recruiter's routine work goes through and an upload of candidate files to an unknown site waits, because Runtime judges each action in context.
Each decision uses the identity, tool, destination, repository, and policy behind the action.
Decide with context
Keep routine work moving and stop clear violations. Runtime allows routine work, warns when risk is explainable, requests approval for ambiguity, and blocks clear policy violations.
Keep people in flow
People stay in their chat or coding session when Runtime steps in, whether they use the Claude desktop app or Claude Code. It explains the matched rule and offers a safe next step right there.
How it works
From requested action to explainable decision.
- 01Observe
A hook, gateway, proxy, extension, or API captures the proposed action.
- 02Enrich
Signals Corps adds user, agent, repository, destination, and capability context.
- 03Decide
The shared policy returns allow, warn, approve, or block with matched evidence.
- 04Record
The employee gets an immediate explanation while IT & Security teams get the audit trail.
Session
agent:research-assistant · Claude Code actions- Delete a file outside the workspace
File system write · high · Blocked
- Read a file in the workspace
Allowed
- Web search
Allowed
- Agent session started
Policy: Engineering default
Coverage without lock-in
Protect the tools employees already use.
Teams keep their tools, and one policy reaches each one through the deepest control point it exposes. You see the enforcement strength for each tool, so you know where an action can be stopped.
- Claude Code and Codex hooks
- Windsurf, Cursor, and compatible IDE adapters
- Agent frameworks, MCP gateways, and proxies
- Custom agents through the engine API
Claude desktop · HRResumes to unapproved siteNeeds approval
ChatGPT desktop · SalesCustomer sheet to personal emailBlocked
Codex · Engineeringgit push --forceBlocked
Cursor · Webnpm testAllowed
Runtime capabilities
Move from visibility to enforcement safely.
Policy simulation
Roll out without surprises by previewing decisions against real activity before they affect anyone.
Approval workflows
Let the right owner approve an unusual action, with its context intact.
Secret boundaries
Stop credentials and sensitive files moving to untrusted tools or destinations.
Local decisions
Get fast feedback and keep control where cloud connectivity is constrained. Prompts and file paths stay on the device, and a command reaches Control Plane only when a rule asks someone to approve it.
Session timelines
Answer what happened with the actions, policies, approvals, and outcomes behind an event, in order.
Custom policy
Enforce your own rules everywhere. Write organization-specific controls once and apply them across every integration.
Headers, llms.txt, and robots.txt compared: what made AI agents act
One sign-up instruction in a header, llms.txt, and robots.txt. Under a plain prompt no Claude agent signed up; granting the site authority moved Sonnet and Haiku, not Opus or robots.txt.
Team Threat Research
Signals Corps Control Plane
Reduce risk across every stage from one security platform.
Set policy, see coverage, stop high-risk actions, and investigate incidents across every agent and team from one operational view.
Write a rule once and have each control point apply it.
Know who is behind every agent, with users, devices, teams, and tools connected.
Defend every decision, with the rule, relevant details, and outcome kept together.
Roll out with one endpoint agent, then manage exceptions, approvals, and reporting in one place.
After agents act
Investigate what the agent did.
Turn agent logs from every team into investigation-ready evidence and find suspicious AI activity fast.