Runtime¶

ClayRuntime is the secure execution wrapper. It fires shields before input reaches your function and after the output is returned — automatically.

Basic Usage¶

from openclay import ClayRuntime, StrictPolicy

runtime = ClayRuntime(policy=StrictPolicy())

result = runtime.run(
    my_llm_function,
    "Analyze this data",
    context="You are a research assistant.",
)

if result.blocked:
    print(f"Blocked at layer: {result.trace.layer}")
    print(f"Reason: {result.trace.reason}")
else:
    print(result.output)

ClayResult¶

Every runtime.run() returns a ClayResult:

Field	Type	Description
`output`	`Any`	The function's return value (or `None` if blocked)
`blocked`	`bool`	Whether the execution was blocked
`trace`	`Trace`	Full explainability record
`elapsed`	`float`	Execution time in seconds

Wrapping Existing Agents¶

Drop-in shielding for LangChain, CrewAI, or any callable:

wrapped = runtime.wrap(my_langchain_agent)
result = wrapped.run("Research AI security trends")

Disabling Layers¶

Temporarily bypass specific shield layers:

with runtime.disable("input"):
    # Input shield is skipped; output shield still fires
    result = runtime.run(my_fn, trusted_input)

Policy Integration¶

The runtime's behaviour is controlled by a Policy:

from openclay import ClayRuntime, AuditPolicy

# Audit mode: scans everything, blocks nothing
runtime = ClayRuntime(policy=AuditPolicy())

result = runtime.run(my_fn, user_input)
# result.blocked is always False in audit mode
# result.trace still contains full shield analysis