Agent runtime
Planning, model calls, memory, and retries
Requests a tool action with an actor and target.
Execution path
Put policy between the agent and production.
The agent runtime plans the action. Identity sets standing access. Deixic checks workspace policy before the tool call and gets approval for sensitive changes.
System boundary
How an agent action gets a decision.
Agent frameworksModel providersSupported systemsMCP tools
Deixic
Allow this agent action?identity + ownertarget + scopepolicyhuman approval
Allow, deny, or waitConnected-system resultActivity historyAgent cost
Execution path
One action across the stack
Each system retains authority over its own data. Deixic owns the policy and approval state for the action it receives.
Identity and credentials
Standing access and credential scope
Supplies the identity and credential reference used for the request.
Deixic
Workspace policy, approval state, and action history
Allows, denies, or pauses the action before execution.
Connected system
Business data and the resulting side effect
Executes an accepted call and returns the result.
Security and compliance systems
Security findings, controls, audits, and compliance evidence
Remain authoritative for their own findings and control status.
Start point
Choose an agent action that changes a production system.
Start with one production action that lacks a defined approval path.
See the product · Browse supported connectors · Request a demo