Protocol Adapter

NativeTacticAdapter exposes a native object as a protocol tactic. The service layer sees TacticInfo; native prompts, dialogs, sessions, and invokers stay behind the adapter.

Expose A Native Tactic

from lllm.runtimes.native import NativeTacticAdapter

public_tactic = NativeTacticAdapter(
    native_tactic,
    package_ref="psi://demo/native/tactics/brief",
)

The adapter preserves the protocol boundary: callers send typed input and receive typed output, while native code can still use agent groups, dialogs, handlers, and sessions internally.

Runtime-Owned Surroundings

The adapter can pass runtime-owned kwargs and call metadata without promoting them into protocol concepts.

from lllm import CallContext
from lllm.runtimes.native import NativeTacticAdapter

tactic = NativeTacticAdapter(
    native_tactic,
    run_kwargs={"tone": "precise"},
)

result = tactic.run(
    {"topic": "native services"},
    context=CallContext(trace_id="trace-1", metadata={"caller": "demo"}),
)

If the native method accepts context, it receives the CallContext. If it accepts metadata, the adapter adds safe LLLM metadata such as request id, trace id, refs, endpoint, and tags.

Tactics As Tools

Native can consume a protocol tactic as a prompt tool:

from lllm.runtimes.native import Prompt, tactic_as_function

lookup = tactic_as_function(search_tactic, parameter_mode="kwargs")
prompt = Prompt(
    path="research/answer",
    prompt="Use tools when needed.",
    function_list=[lookup],
)

Native tactics can also expose ordinary callable tools for Pydantic AI and other runtimes:

tool_callable = tactic.as_tool(parameter_mode="task")

For methods on a native tactic class, use @tactictool when a method should be discoverable as a native tool resource. Selector refs can use #tool_name to refer to a method-level tool.