Executor boundary
Wrap the tool or action executor that performs consequential work. Fits Amazon Bedrock Agents, custom runtimes, APIs, containers, Kubernetes, and VM-hosted agents.
Architecture
Built for enforcement in the execution path.
Tegrix attaches at the boundary where agent actions already pass. No agent rewrites, no platform replacement, and no separate log reconstruction after the fact.
Executor, gateway, MCP, and interceptor attachment patterns.
Local boundary decisions with cached authority and fail-safe policy.
Evidence recorded when the decision is made, before execution continues.
Execution-path control
Decision at the boundary. Evidence at the moment.
1Agent attempts action
2Runtime boundary intercepts
3Authority Envelope loads
4Facts and controls evaluate
5Decision returns locally
6Provider executes or stops
7Evidence is recorded
How Tegrix attaches
The action path is the control point.
Boundary placement makes governance enforceable. Registration establishes identity and association; the runtime boundary enforces the decision before action.
Gateway boundary
Intercept the tool call where the platform already routes agent actions. Fits Amazon Bedrock AgentCore gateways and cloud gateway extension paths.
MCP / interceptor boundary
Govern MCP servers, internal APIs, service workflows, and automation paths before the action reaches the business system.
Different runtimes. One authority model.
Runtime behavior
Decisions run where execution is about to happen.
The runtime boundary evaluates the Authority Envelope, facts, controls, approvals, and stop state before the provider or business system receives the action.
Local evaluation
A decision bundle and cached Authority Envelope are evaluated at the runtime boundary itself, so routine decisions do not require a round-trip to Tegrix.
Measured overhead
Current live paths measure approximately 80-110ms p50 decision overhead before provider execution continues.
Fail-safe posture
Consequential paths fail safe by policy when the required bundle, facts, approval state, or emergency-stop signal is unavailable.
Emergency stop
Stop state is replicated to governed runtime boundaries in seconds and uses the same evidence path as other decisions.
Data boundary
Enough context to decide. Not a copy of your systems.
Tegrix uses decision fields and evidence metadata from systems you already run. Source systems remain authoritative.
Agent identity
Provider identity, runtime identity, service principal, owner, environment, and assigned Authority Envelope.
Action request
The attempted action, tool, resource, scope, requested value, target system, and correlation identifiers.
Enterprise facts
Only the facts needed for the decision, such as identity, approval status, change freeze, maintenance window, threshold, and freshness.
Control result
Matched controls, decision, human approval state when required, provider invocation state, and final execution outcome.
Evidence and bypass
Every decision leaves evidence from the decision path.
The Evidence Ledger captures what was attempted, which authority and facts were evaluated, what Tegrix decided, and what executed.
Not a log replay
Because Tegrix makes the decision, evidence is recorded at the moment of decision, not reconstructed from application logs afterward.
Bypass boundary
The action path is the control point. If a consequential action does not pass a governed boundary, registration alone does not govern it.
Provider continuity
AWS, Microsoft, SaaS, and internal systems continue to operate in place while Tegrix applies the same authority model at the boundary.
Security review
The first questions before inline enforcement.
What do I deploy?
An attachment at the executor, gateway, MCP, or interceptor boundary where the agent already attempts action.
Do I rewrite agents?
No. Tegrix attaches where agents already act; platforms, tools, and business systems remain in place.
What if Tegrix is down?
Runtime boundaries evaluate cached authority locally and fail safe by policy on consequential paths.
What data crosses?
Decision fields and evidence metadata cross, not full source-system copies.
Can an agent bypass it?
Only actions routed outside a governed boundary avoid enforcement, which is why boundary mapping is the first architecture step.
Why trust the ledger?
The evidence record is written from the same boundary decision that allows, denies, approves, or stops execution.
Bring one consequential workflow. We map the agent, boundary, Authority Envelope, fail-safe posture, and evidence record.
Production systems stay authoritative.