Skip to content

Middleware

Framework adapters for LangChain, CrewAI, AutoGen, Claude ADK, and Google ADK.

middleware

Framework adapter interfaces.

SafeAIAutoGenAdapter

Bases: BaseMiddleware

Adapter that wraps AutoGen tool execution paths.

Source code in safeai/middleware/autogen.py 
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
class SafeAIAutoGenAdapter(BaseMiddleware):
    """Adapter that wraps AutoGen tool execution paths."""

    def wrap_tool(
        self,
        tool_name: str,
        fn: Callable[..., Any],
        *,
        agent_id: str = "unknown",
        session_id: str | None = None,
        source_agent_id: str | None = None,
        destination_agent_id: str | None = None,
        request_data_tags: list[str] | None = None,
        capability_token_id: str | None = None,
        capability_action: str = "invoke",
        approval_request_id: str | None = None,
    ) -> Callable[..., Any]:
        """Wrap a synchronous AutoGen tool callable."""

        @wraps(fn)
        def _wrapped(*args: Any, **kwargs: Any) -> Any:
            payload, shape = _normalize_input(args, kwargs)
            tags = list(request_data_tags or [])
            request = self.safeai.intercept_tool_request(
                tool_name=tool_name,
                parameters=payload,
                data_tags=tags,
                agent_id=agent_id,
                session_id=session_id,
                source_agent_id=source_agent_id,
                destination_agent_id=destination_agent_id,
                action_type="autogen_tool",
                capability_token_id=capability_token_id,
                capability_action=capability_action,
                approval_request_id=approval_request_id,
            )
            if request.decision.action != "allow":
                raise SafeAIBlockedError(
                    action=request.decision.action,
                    policy_name=request.decision.policy_name,
                    reason=request.decision.reason,
                )

            result = _invoke_with_shape(fn, request.filtered_params, shape)
            guarded = self.safeai.intercept_tool_response(
                tool_name=tool_name,
                response=_normalize_response(result),
                agent_id=agent_id,
                request_data_tags=tags,
                session_id=session_id,
                source_agent_id=source_agent_id,
                destination_agent_id=destination_agent_id,
                action_type="autogen_tool",
            )
            if guarded.decision.action in {"block", "require_approval"}:
                raise SafeAIBlockedError(
                    action=guarded.decision.action,
                    policy_name=guarded.decision.policy_name,
                    reason=guarded.decision.reason,
                )
            return _restore_response_shape(result, guarded.filtered_response)

        return _wrapped

    def wrap_async_tool(
        self,
        tool_name: str,
        fn: Callable[..., Any],
        *,
        agent_id: str = "unknown",
        session_id: str | None = None,
        source_agent_id: str | None = None,
        destination_agent_id: str | None = None,
        request_data_tags: list[str] | None = None,
        capability_token_id: str | None = None,
        capability_action: str = "invoke",
        approval_request_id: str | None = None,
    ) -> Callable[..., Any]:
        """Wrap an asynchronous AutoGen tool callable."""

        @wraps(fn)
        async def _wrapped(*args: Any, **kwargs: Any) -> Any:
            payload, shape = _normalize_input(args, kwargs)
            tags = list(request_data_tags or [])
            request = self.safeai.intercept_tool_request(
                tool_name=tool_name,
                parameters=payload,
                data_tags=tags,
                agent_id=agent_id,
                session_id=session_id,
                source_agent_id=source_agent_id,
                destination_agent_id=destination_agent_id,
                action_type="autogen_tool",
                capability_token_id=capability_token_id,
                capability_action=capability_action,
                approval_request_id=approval_request_id,
            )
            if request.decision.action != "allow":
                raise SafeAIBlockedError(
                    action=request.decision.action,
                    policy_name=request.decision.policy_name,
                    reason=request.decision.reason,
                )

            result = await _ainvoke_with_shape(fn, request.filtered_params, shape)
            guarded = self.safeai.intercept_tool_response(
                tool_name=tool_name,
                response=_normalize_response(result),
                agent_id=agent_id,
                request_data_tags=tags,
                session_id=session_id,
                source_agent_id=source_agent_id,
                destination_agent_id=destination_agent_id,
                action_type="autogen_tool",
            )
            if guarded.decision.action in {"block", "require_approval"}:
                raise SafeAIBlockedError(
                    action=guarded.decision.action,
                    policy_name=guarded.decision.policy_name,
                    reason=guarded.decision.reason,
                )
            return _restore_response_shape(result, guarded.filtered_response)

        return _wrapped

wrap_tool 

wrap_tool(tool_name: str, fn: Callable[..., Any], *, agent_id: str = 'unknown', session_id: str | None = None, source_agent_id: str | None = None, destination_agent_id: str | None = None, request_data_tags: list[str] | None = None, capability_token_id: str | None = None, capability_action: str = 'invoke', approval_request_id: str | None = None) -> Callable[..., Any]

Wrap a synchronous AutoGen tool callable.

Source code in safeai/middleware/autogen.py 
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
def wrap_tool(
    self,
    tool_name: str,
    fn: Callable[..., Any],
    *,
    agent_id: str = "unknown",
    session_id: str | None = None,
    source_agent_id: str | None = None,
    destination_agent_id: str | None = None,
    request_data_tags: list[str] | None = None,
    capability_token_id: str | None = None,
    capability_action: str = "invoke",
    approval_request_id: str | None = None,
) -> Callable[..., Any]:
    """Wrap a synchronous AutoGen tool callable."""

    @wraps(fn)
    def _wrapped(*args: Any, **kwargs: Any) -> Any:
        payload, shape = _normalize_input(args, kwargs)
        tags = list(request_data_tags or [])
        request = self.safeai.intercept_tool_request(
            tool_name=tool_name,
            parameters=payload,
            data_tags=tags,
            agent_id=agent_id,
            session_id=session_id,
            source_agent_id=source_agent_id,
            destination_agent_id=destination_agent_id,
            action_type="autogen_tool",
            capability_token_id=capability_token_id,
            capability_action=capability_action,
            approval_request_id=approval_request_id,
        )
        if request.decision.action != "allow":
            raise SafeAIBlockedError(
                action=request.decision.action,
                policy_name=request.decision.policy_name,
                reason=request.decision.reason,
            )

        result = _invoke_with_shape(fn, request.filtered_params, shape)
        guarded = self.safeai.intercept_tool_response(
            tool_name=tool_name,
            response=_normalize_response(result),
            agent_id=agent_id,
            request_data_tags=tags,
            session_id=session_id,
            source_agent_id=source_agent_id,
            destination_agent_id=destination_agent_id,
            action_type="autogen_tool",
        )
        if guarded.decision.action in {"block", "require_approval"}:
            raise SafeAIBlockedError(
                action=guarded.decision.action,
                policy_name=guarded.decision.policy_name,
                reason=guarded.decision.reason,
            )
        return _restore_response_shape(result, guarded.filtered_response)

    return _wrapped

wrap_async_tool 

wrap_async_tool(tool_name: str, fn: Callable[..., Any], *, agent_id: str = 'unknown', session_id: str | None = None, source_agent_id: str | None = None, destination_agent_id: str | None = None, request_data_tags: list[str] | None = None, capability_token_id: str | None = None, capability_action: str = 'invoke', approval_request_id: str | None = None) -> Callable[..., Any]

Wrap an asynchronous AutoGen tool callable.

Source code in safeai/middleware/autogen.py 
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
def wrap_async_tool(
    self,
    tool_name: str,
    fn: Callable[..., Any],
    *,
    agent_id: str = "unknown",
    session_id: str | None = None,
    source_agent_id: str | None = None,
    destination_agent_id: str | None = None,
    request_data_tags: list[str] | None = None,
    capability_token_id: str | None = None,
    capability_action: str = "invoke",
    approval_request_id: str | None = None,
) -> Callable[..., Any]:
    """Wrap an asynchronous AutoGen tool callable."""

    @wraps(fn)
    async def _wrapped(*args: Any, **kwargs: Any) -> Any:
        payload, shape = _normalize_input(args, kwargs)
        tags = list(request_data_tags or [])
        request = self.safeai.intercept_tool_request(
            tool_name=tool_name,
            parameters=payload,
            data_tags=tags,
            agent_id=agent_id,
            session_id=session_id,
            source_agent_id=source_agent_id,
            destination_agent_id=destination_agent_id,
            action_type="autogen_tool",
            capability_token_id=capability_token_id,
            capability_action=capability_action,
            approval_request_id=approval_request_id,
        )
        if request.decision.action != "allow":
            raise SafeAIBlockedError(
                action=request.decision.action,
                policy_name=request.decision.policy_name,
                reason=request.decision.reason,
            )

        result = await _ainvoke_with_shape(fn, request.filtered_params, shape)
        guarded = self.safeai.intercept_tool_response(
            tool_name=tool_name,
            response=_normalize_response(result),
            agent_id=agent_id,
            request_data_tags=tags,
            session_id=session_id,
            source_agent_id=source_agent_id,
            destination_agent_id=destination_agent_id,
            action_type="autogen_tool",
        )
        if guarded.decision.action in {"block", "require_approval"}:
            raise SafeAIBlockedError(
                action=guarded.decision.action,
                policy_name=guarded.decision.policy_name,
                reason=guarded.decision.reason,
            )
        return _restore_response_shape(result, guarded.filtered_response)

    return _wrapped

SafeAIClaudeADKAdapter

Bases: BaseMiddleware

Adapter that wraps Claude ADK tool execution paths.

Source code in safeai/middleware/claude_adk.py 
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
class SafeAIClaudeADKAdapter(BaseMiddleware):
    """Adapter that wraps Claude ADK tool execution paths."""

    def wrap_tool(
        self,
        tool_name: str,
        fn: Callable[..., Any],
        *,
        agent_id: str = "unknown",
        session_id: str | None = None,
        source_agent_id: str | None = None,
        destination_agent_id: str | None = None,
        request_data_tags: list[str] | None = None,
        capability_token_id: str | None = None,
        capability_action: str = "invoke",
        approval_request_id: str | None = None,
    ) -> Callable[..., Any]:
        """Wrap a synchronous Claude ADK tool callable."""

        @wraps(fn)
        def _wrapped(*args: Any, **kwargs: Any) -> Any:
            payload, shape = _normalize_input(args, kwargs)
            tags = list(request_data_tags or [])

            request = self.safeai.intercept_tool_request(
                tool_name=tool_name,
                parameters=payload,
                data_tags=tags,
                agent_id=agent_id,
                session_id=session_id,
                source_agent_id=source_agent_id,
                destination_agent_id=destination_agent_id,
                action_type="claude_adk_tool",
                capability_token_id=capability_token_id,
                capability_action=capability_action,
                approval_request_id=approval_request_id,
            )
            if request.decision.action != "allow":
                raise SafeAIBlockedError(
                    action=request.decision.action,
                    policy_name=request.decision.policy_name,
                    reason=request.decision.reason,
                )

            result = _invoke_with_shape(fn, request.filtered_params, shape)
            guarded = self.safeai.intercept_tool_response(
                tool_name=tool_name,
                response=_normalize_response(result),
                agent_id=agent_id,
                request_data_tags=tags,
                session_id=session_id,
                source_agent_id=source_agent_id,
                destination_agent_id=destination_agent_id,
                action_type="claude_adk_tool",
            )
            if guarded.decision.action in {"block", "require_approval"}:
                raise SafeAIBlockedError(
                    action=guarded.decision.action,
                    policy_name=guarded.decision.policy_name,
                    reason=guarded.decision.reason,
                )
            return _restore_response_shape(result, guarded.filtered_response)

        return _wrapped

    def wrap_async_tool(
        self,
        tool_name: str,
        fn: Callable[..., Any],
        *,
        agent_id: str = "unknown",
        session_id: str | None = None,
        source_agent_id: str | None = None,
        destination_agent_id: str | None = None,
        request_data_tags: list[str] | None = None,
        capability_token_id: str | None = None,
        capability_action: str = "invoke",
        approval_request_id: str | None = None,
    ) -> Callable[..., Any]:
        """Wrap an asynchronous Claude ADK tool callable."""

        @wraps(fn)
        async def _wrapped(*args: Any, **kwargs: Any) -> Any:
            payload, shape = _normalize_input(args, kwargs)
            tags = list(request_data_tags or [])
            request = self.safeai.intercept_tool_request(
                tool_name=tool_name,
                parameters=payload,
                data_tags=tags,
                agent_id=agent_id,
                session_id=session_id,
                source_agent_id=source_agent_id,
                destination_agent_id=destination_agent_id,
                action_type="claude_adk_tool",
                capability_token_id=capability_token_id,
                capability_action=capability_action,
                approval_request_id=approval_request_id,
            )
            if request.decision.action != "allow":
                raise SafeAIBlockedError(
                    action=request.decision.action,
                    policy_name=request.decision.policy_name,
                    reason=request.decision.reason,
                )

            result = await _ainvoke_with_shape(fn, request.filtered_params, shape)
            guarded = self.safeai.intercept_tool_response(
                tool_name=tool_name,
                response=_normalize_response(result),
                agent_id=agent_id,
                request_data_tags=tags,
                session_id=session_id,
                source_agent_id=source_agent_id,
                destination_agent_id=destination_agent_id,
                action_type="claude_adk_tool",
            )
            if guarded.decision.action in {"block", "require_approval"}:
                raise SafeAIBlockedError(
                    action=guarded.decision.action,
                    policy_name=guarded.decision.policy_name,
                    reason=guarded.decision.reason,
                )
            return _restore_response_shape(result, guarded.filtered_response)

        return _wrapped

wrap_tool 

wrap_tool(tool_name: str, fn: Callable[..., Any], *, agent_id: str = 'unknown', session_id: str | None = None, source_agent_id: str | None = None, destination_agent_id: str | None = None, request_data_tags: list[str] | None = None, capability_token_id: str | None = None, capability_action: str = 'invoke', approval_request_id: str | None = None) -> Callable[..., Any]

Wrap a synchronous Claude ADK tool callable.

Source code in safeai/middleware/claude_adk.py 
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
def wrap_tool(
    self,
    tool_name: str,
    fn: Callable[..., Any],
    *,
    agent_id: str = "unknown",
    session_id: str | None = None,
    source_agent_id: str | None = None,
    destination_agent_id: str | None = None,
    request_data_tags: list[str] | None = None,
    capability_token_id: str | None = None,
    capability_action: str = "invoke",
    approval_request_id: str | None = None,
) -> Callable[..., Any]:
    """Wrap a synchronous Claude ADK tool callable."""

    @wraps(fn)
    def _wrapped(*args: Any, **kwargs: Any) -> Any:
        payload, shape = _normalize_input(args, kwargs)
        tags = list(request_data_tags or [])

        request = self.safeai.intercept_tool_request(
            tool_name=tool_name,
            parameters=payload,
            data_tags=tags,
            agent_id=agent_id,
            session_id=session_id,
            source_agent_id=source_agent_id,
            destination_agent_id=destination_agent_id,
            action_type="claude_adk_tool",
            capability_token_id=capability_token_id,
            capability_action=capability_action,
            approval_request_id=approval_request_id,
        )
        if request.decision.action != "allow":
            raise SafeAIBlockedError(
                action=request.decision.action,
                policy_name=request.decision.policy_name,
                reason=request.decision.reason,
            )

        result = _invoke_with_shape(fn, request.filtered_params, shape)
        guarded = self.safeai.intercept_tool_response(
            tool_name=tool_name,
            response=_normalize_response(result),
            agent_id=agent_id,
            request_data_tags=tags,
            session_id=session_id,
            source_agent_id=source_agent_id,
            destination_agent_id=destination_agent_id,
            action_type="claude_adk_tool",
        )
        if guarded.decision.action in {"block", "require_approval"}:
            raise SafeAIBlockedError(
                action=guarded.decision.action,
                policy_name=guarded.decision.policy_name,
                reason=guarded.decision.reason,
            )
        return _restore_response_shape(result, guarded.filtered_response)

    return _wrapped

wrap_async_tool 

wrap_async_tool(tool_name: str, fn: Callable[..., Any], *, agent_id: str = 'unknown', session_id: str | None = None, source_agent_id: str | None = None, destination_agent_id: str | None = None, request_data_tags: list[str] | None = None, capability_token_id: str | None = None, capability_action: str = 'invoke', approval_request_id: str | None = None) -> Callable[..., Any]

Wrap an asynchronous Claude ADK tool callable.

Source code in safeai/middleware/claude_adk.py 
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
def wrap_async_tool(
    self,
    tool_name: str,
    fn: Callable[..., Any],
    *,
    agent_id: str = "unknown",
    session_id: str | None = None,
    source_agent_id: str | None = None,
    destination_agent_id: str | None = None,
    request_data_tags: list[str] | None = None,
    capability_token_id: str | None = None,
    capability_action: str = "invoke",
    approval_request_id: str | None = None,
) -> Callable[..., Any]:
    """Wrap an asynchronous Claude ADK tool callable."""

    @wraps(fn)
    async def _wrapped(*args: Any, **kwargs: Any) -> Any:
        payload, shape = _normalize_input(args, kwargs)
        tags = list(request_data_tags or [])
        request = self.safeai.intercept_tool_request(
            tool_name=tool_name,
            parameters=payload,
            data_tags=tags,
            agent_id=agent_id,
            session_id=session_id,
            source_agent_id=source_agent_id,
            destination_agent_id=destination_agent_id,
            action_type="claude_adk_tool",
            capability_token_id=capability_token_id,
            capability_action=capability_action,
            approval_request_id=approval_request_id,
        )
        if request.decision.action != "allow":
            raise SafeAIBlockedError(
                action=request.decision.action,
                policy_name=request.decision.policy_name,
                reason=request.decision.reason,
            )

        result = await _ainvoke_with_shape(fn, request.filtered_params, shape)
        guarded = self.safeai.intercept_tool_response(
            tool_name=tool_name,
            response=_normalize_response(result),
            agent_id=agent_id,
            request_data_tags=tags,
            session_id=session_id,
            source_agent_id=source_agent_id,
            destination_agent_id=destination_agent_id,
            action_type="claude_adk_tool",
        )
        if guarded.decision.action in {"block", "require_approval"}:
            raise SafeAIBlockedError(
                action=guarded.decision.action,
                policy_name=guarded.decision.policy_name,
                reason=guarded.decision.reason,
            )
        return _restore_response_shape(result, guarded.filtered_response)

    return _wrapped

SafeAICrewAIAdapter

Bases: BaseMiddleware

Adapter that wraps CrewAI tool execution paths.

Source code in safeai/middleware/crewai.py 
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
class SafeAICrewAIAdapter(BaseMiddleware):
    """Adapter that wraps CrewAI tool execution paths."""

    def wrap_tool(
        self,
        tool_name: str,
        fn: Callable[..., Any],
        *,
        agent_id: str = "unknown",
        session_id: str | None = None,
        source_agent_id: str | None = None,
        destination_agent_id: str | None = None,
        request_data_tags: list[str] | None = None,
        capability_token_id: str | None = None,
        capability_action: str = "invoke",
        approval_request_id: str | None = None,
    ) -> Callable[..., Any]:
        """Wrap a synchronous CrewAI tool callable."""

        @wraps(fn)
        def _wrapped(*args: Any, **kwargs: Any) -> Any:
            payload, shape = _normalize_input(args, kwargs)
            tags = list(request_data_tags or [])
            request = self.safeai.intercept_tool_request(
                tool_name=tool_name,
                parameters=payload,
                data_tags=tags,
                agent_id=agent_id,
                session_id=session_id,
                source_agent_id=source_agent_id,
                destination_agent_id=destination_agent_id,
                action_type="crewai_tool",
                capability_token_id=capability_token_id,
                capability_action=capability_action,
                approval_request_id=approval_request_id,
            )
            if request.decision.action != "allow":
                raise SafeAIBlockedError(
                    action=request.decision.action,
                    policy_name=request.decision.policy_name,
                    reason=request.decision.reason,
                )

            result = _invoke_with_shape(fn, request.filtered_params, shape)
            guarded = self.safeai.intercept_tool_response(
                tool_name=tool_name,
                response=_normalize_response(result),
                agent_id=agent_id,
                request_data_tags=tags,
                session_id=session_id,
                source_agent_id=source_agent_id,
                destination_agent_id=destination_agent_id,
                action_type="crewai_tool",
            )
            if guarded.decision.action in {"block", "require_approval"}:
                raise SafeAIBlockedError(
                    action=guarded.decision.action,
                    policy_name=guarded.decision.policy_name,
                    reason=guarded.decision.reason,
                )
            return _restore_response_shape(result, guarded.filtered_response)

        return _wrapped

    def wrap_async_tool(
        self,
        tool_name: str,
        fn: Callable[..., Any],
        *,
        agent_id: str = "unknown",
        session_id: str | None = None,
        source_agent_id: str | None = None,
        destination_agent_id: str | None = None,
        request_data_tags: list[str] | None = None,
        capability_token_id: str | None = None,
        capability_action: str = "invoke",
        approval_request_id: str | None = None,
    ) -> Callable[..., Any]:
        """Wrap an asynchronous CrewAI tool callable."""

        @wraps(fn)
        async def _wrapped(*args: Any, **kwargs: Any) -> Any:
            payload, shape = _normalize_input(args, kwargs)
            tags = list(request_data_tags or [])
            request = self.safeai.intercept_tool_request(
                tool_name=tool_name,
                parameters=payload,
                data_tags=tags,
                agent_id=agent_id,
                session_id=session_id,
                source_agent_id=source_agent_id,
                destination_agent_id=destination_agent_id,
                action_type="crewai_tool",
                capability_token_id=capability_token_id,
                capability_action=capability_action,
                approval_request_id=approval_request_id,
            )
            if request.decision.action != "allow":
                raise SafeAIBlockedError(
                    action=request.decision.action,
                    policy_name=request.decision.policy_name,
                    reason=request.decision.reason,
                )

            result = await _ainvoke_with_shape(fn, request.filtered_params, shape)
            guarded = self.safeai.intercept_tool_response(
                tool_name=tool_name,
                response=_normalize_response(result),
                agent_id=agent_id,
                request_data_tags=tags,
                session_id=session_id,
                source_agent_id=source_agent_id,
                destination_agent_id=destination_agent_id,
                action_type="crewai_tool",
            )
            if guarded.decision.action in {"block", "require_approval"}:
                raise SafeAIBlockedError(
                    action=guarded.decision.action,
                    policy_name=guarded.decision.policy_name,
                    reason=guarded.decision.reason,
                )
            return _restore_response_shape(result, guarded.filtered_response)

        return _wrapped

wrap_tool 

wrap_tool(tool_name: str, fn: Callable[..., Any], *, agent_id: str = 'unknown', session_id: str | None = None, source_agent_id: str | None = None, destination_agent_id: str | None = None, request_data_tags: list[str] | None = None, capability_token_id: str | None = None, capability_action: str = 'invoke', approval_request_id: str | None = None) -> Callable[..., Any]

Wrap a synchronous CrewAI tool callable.

Source code in safeai/middleware/crewai.py 
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
def wrap_tool(
    self,
    tool_name: str,
    fn: Callable[..., Any],
    *,
    agent_id: str = "unknown",
    session_id: str | None = None,
    source_agent_id: str | None = None,
    destination_agent_id: str | None = None,
    request_data_tags: list[str] | None = None,
    capability_token_id: str | None = None,
    capability_action: str = "invoke",
    approval_request_id: str | None = None,
) -> Callable[..., Any]:
    """Wrap a synchronous CrewAI tool callable."""

    @wraps(fn)
    def _wrapped(*args: Any, **kwargs: Any) -> Any:
        payload, shape = _normalize_input(args, kwargs)
        tags = list(request_data_tags or [])
        request = self.safeai.intercept_tool_request(
            tool_name=tool_name,
            parameters=payload,
            data_tags=tags,
            agent_id=agent_id,
            session_id=session_id,
            source_agent_id=source_agent_id,
            destination_agent_id=destination_agent_id,
            action_type="crewai_tool",
            capability_token_id=capability_token_id,
            capability_action=capability_action,
            approval_request_id=approval_request_id,
        )
        if request.decision.action != "allow":
            raise SafeAIBlockedError(
                action=request.decision.action,
                policy_name=request.decision.policy_name,
                reason=request.decision.reason,
            )

        result = _invoke_with_shape(fn, request.filtered_params, shape)
        guarded = self.safeai.intercept_tool_response(
            tool_name=tool_name,
            response=_normalize_response(result),
            agent_id=agent_id,
            request_data_tags=tags,
            session_id=session_id,
            source_agent_id=source_agent_id,
            destination_agent_id=destination_agent_id,
            action_type="crewai_tool",
        )
        if guarded.decision.action in {"block", "require_approval"}:
            raise SafeAIBlockedError(
                action=guarded.decision.action,
                policy_name=guarded.decision.policy_name,
                reason=guarded.decision.reason,
            )
        return _restore_response_shape(result, guarded.filtered_response)

    return _wrapped

wrap_async_tool 

wrap_async_tool(tool_name: str, fn: Callable[..., Any], *, agent_id: str = 'unknown', session_id: str | None = None, source_agent_id: str | None = None, destination_agent_id: str | None = None, request_data_tags: list[str] | None = None, capability_token_id: str | None = None, capability_action: str = 'invoke', approval_request_id: str | None = None) -> Callable[..., Any]

Wrap an asynchronous CrewAI tool callable.

Source code in safeai/middleware/crewai.py 
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
def wrap_async_tool(
    self,
    tool_name: str,
    fn: Callable[..., Any],
    *,
    agent_id: str = "unknown",
    session_id: str | None = None,
    source_agent_id: str | None = None,
    destination_agent_id: str | None = None,
    request_data_tags: list[str] | None = None,
    capability_token_id: str | None = None,
    capability_action: str = "invoke",
    approval_request_id: str | None = None,
) -> Callable[..., Any]:
    """Wrap an asynchronous CrewAI tool callable."""

    @wraps(fn)
    async def _wrapped(*args: Any, **kwargs: Any) -> Any:
        payload, shape = _normalize_input(args, kwargs)
        tags = list(request_data_tags or [])
        request = self.safeai.intercept_tool_request(
            tool_name=tool_name,
            parameters=payload,
            data_tags=tags,
            agent_id=agent_id,
            session_id=session_id,
            source_agent_id=source_agent_id,
            destination_agent_id=destination_agent_id,
            action_type="crewai_tool",
            capability_token_id=capability_token_id,
            capability_action=capability_action,
            approval_request_id=approval_request_id,
        )
        if request.decision.action != "allow":
            raise SafeAIBlockedError(
                action=request.decision.action,
                policy_name=request.decision.policy_name,
                reason=request.decision.reason,
            )

        result = await _ainvoke_with_shape(fn, request.filtered_params, shape)
        guarded = self.safeai.intercept_tool_response(
            tool_name=tool_name,
            response=_normalize_response(result),
            agent_id=agent_id,
            request_data_tags=tags,
            session_id=session_id,
            source_agent_id=source_agent_id,
            destination_agent_id=destination_agent_id,
            action_type="crewai_tool",
        )
        if guarded.decision.action in {"block", "require_approval"}:
            raise SafeAIBlockedError(
                action=guarded.decision.action,
                policy_name=guarded.decision.policy_name,
                reason=guarded.decision.reason,
            )
        return _restore_response_shape(result, guarded.filtered_response)

    return _wrapped

SafeAIGoogleADKAdapter

Bases: BaseMiddleware

Adapter that wraps Google ADK tool execution paths.

Source code in safeai/middleware/google_adk.py 
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
class SafeAIGoogleADKAdapter(BaseMiddleware):
    """Adapter that wraps Google ADK tool execution paths."""

    def wrap_tool(
        self,
        tool_name: str,
        fn: Callable[..., Any],
        *,
        agent_id: str = "unknown",
        session_id: str | None = None,
        source_agent_id: str | None = None,
        destination_agent_id: str | None = None,
        request_data_tags: list[str] | None = None,
        capability_token_id: str | None = None,
        capability_action: str = "invoke",
        approval_request_id: str | None = None,
    ) -> Callable[..., Any]:
        """Wrap a synchronous Google ADK tool callable."""

        @wraps(fn)
        def _wrapped(*args: Any, **kwargs: Any) -> Any:
            payload, shape = _normalize_input(args, kwargs)
            tags = list(request_data_tags or [])
            request = self.safeai.intercept_tool_request(
                tool_name=tool_name,
                parameters=payload,
                data_tags=tags,
                agent_id=agent_id,
                session_id=session_id,
                source_agent_id=source_agent_id,
                destination_agent_id=destination_agent_id,
                action_type="google_adk_tool",
                capability_token_id=capability_token_id,
                capability_action=capability_action,
                approval_request_id=approval_request_id,
            )
            if request.decision.action != "allow":
                raise SafeAIBlockedError(
                    action=request.decision.action,
                    policy_name=request.decision.policy_name,
                    reason=request.decision.reason,
                )

            result = _invoke_with_shape(fn, request.filtered_params, shape)
            guarded = self.safeai.intercept_tool_response(
                tool_name=tool_name,
                response=_normalize_response(result),
                agent_id=agent_id,
                request_data_tags=tags,
                session_id=session_id,
                source_agent_id=source_agent_id,
                destination_agent_id=destination_agent_id,
                action_type="google_adk_tool",
            )
            if guarded.decision.action in {"block", "require_approval"}:
                raise SafeAIBlockedError(
                    action=guarded.decision.action,
                    policy_name=guarded.decision.policy_name,
                    reason=guarded.decision.reason,
                )
            return _restore_response_shape(result, guarded.filtered_response)

        return _wrapped

    def wrap_async_tool(
        self,
        tool_name: str,
        fn: Callable[..., Any],
        *,
        agent_id: str = "unknown",
        session_id: str | None = None,
        source_agent_id: str | None = None,
        destination_agent_id: str | None = None,
        request_data_tags: list[str] | None = None,
        capability_token_id: str | None = None,
        capability_action: str = "invoke",
        approval_request_id: str | None = None,
    ) -> Callable[..., Any]:
        """Wrap an asynchronous Google ADK tool callable."""

        @wraps(fn)
        async def _wrapped(*args: Any, **kwargs: Any) -> Any:
            payload, shape = _normalize_input(args, kwargs)
            tags = list(request_data_tags or [])
            request = self.safeai.intercept_tool_request(
                tool_name=tool_name,
                parameters=payload,
                data_tags=tags,
                agent_id=agent_id,
                session_id=session_id,
                source_agent_id=source_agent_id,
                destination_agent_id=destination_agent_id,
                action_type="google_adk_tool",
                capability_token_id=capability_token_id,
                capability_action=capability_action,
                approval_request_id=approval_request_id,
            )
            if request.decision.action != "allow":
                raise SafeAIBlockedError(
                    action=request.decision.action,
                    policy_name=request.decision.policy_name,
                    reason=request.decision.reason,
                )

            result = await _ainvoke_with_shape(fn, request.filtered_params, shape)
            guarded = self.safeai.intercept_tool_response(
                tool_name=tool_name,
                response=_normalize_response(result),
                agent_id=agent_id,
                request_data_tags=tags,
                session_id=session_id,
                source_agent_id=source_agent_id,
                destination_agent_id=destination_agent_id,
                action_type="google_adk_tool",
            )
            if guarded.decision.action in {"block", "require_approval"}:
                raise SafeAIBlockedError(
                    action=guarded.decision.action,
                    policy_name=guarded.decision.policy_name,
                    reason=guarded.decision.reason,
                )
            return _restore_response_shape(result, guarded.filtered_response)

        return _wrapped

wrap_tool 

wrap_tool(tool_name: str, fn: Callable[..., Any], *, agent_id: str = 'unknown', session_id: str | None = None, source_agent_id: str | None = None, destination_agent_id: str | None = None, request_data_tags: list[str] | None = None, capability_token_id: str | None = None, capability_action: str = 'invoke', approval_request_id: str | None = None) -> Callable[..., Any]

Wrap a synchronous Google ADK tool callable.

Source code in safeai/middleware/google_adk.py 
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
def wrap_tool(
    self,
    tool_name: str,
    fn: Callable[..., Any],
    *,
    agent_id: str = "unknown",
    session_id: str | None = None,
    source_agent_id: str | None = None,
    destination_agent_id: str | None = None,
    request_data_tags: list[str] | None = None,
    capability_token_id: str | None = None,
    capability_action: str = "invoke",
    approval_request_id: str | None = None,
) -> Callable[..., Any]:
    """Wrap a synchronous Google ADK tool callable."""

    @wraps(fn)
    def _wrapped(*args: Any, **kwargs: Any) -> Any:
        payload, shape = _normalize_input(args, kwargs)
        tags = list(request_data_tags or [])
        request = self.safeai.intercept_tool_request(
            tool_name=tool_name,
            parameters=payload,
            data_tags=tags,
            agent_id=agent_id,
            session_id=session_id,
            source_agent_id=source_agent_id,
            destination_agent_id=destination_agent_id,
            action_type="google_adk_tool",
            capability_token_id=capability_token_id,
            capability_action=capability_action,
            approval_request_id=approval_request_id,
        )
        if request.decision.action != "allow":
            raise SafeAIBlockedError(
                action=request.decision.action,
                policy_name=request.decision.policy_name,
                reason=request.decision.reason,
            )

        result = _invoke_with_shape(fn, request.filtered_params, shape)
        guarded = self.safeai.intercept_tool_response(
            tool_name=tool_name,
            response=_normalize_response(result),
            agent_id=agent_id,
            request_data_tags=tags,
            session_id=session_id,
            source_agent_id=source_agent_id,
            destination_agent_id=destination_agent_id,
            action_type="google_adk_tool",
        )
        if guarded.decision.action in {"block", "require_approval"}:
            raise SafeAIBlockedError(
                action=guarded.decision.action,
                policy_name=guarded.decision.policy_name,
                reason=guarded.decision.reason,
            )
        return _restore_response_shape(result, guarded.filtered_response)

    return _wrapped

wrap_async_tool 

wrap_async_tool(tool_name: str, fn: Callable[..., Any], *, agent_id: str = 'unknown', session_id: str | None = None, source_agent_id: str | None = None, destination_agent_id: str | None = None, request_data_tags: list[str] | None = None, capability_token_id: str | None = None, capability_action: str = 'invoke', approval_request_id: str | None = None) -> Callable[..., Any]

Wrap an asynchronous Google ADK tool callable.

Source code in safeai/middleware/google_adk.py 
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
def wrap_async_tool(
    self,
    tool_name: str,
    fn: Callable[..., Any],
    *,
    agent_id: str = "unknown",
    session_id: str | None = None,
    source_agent_id: str | None = None,
    destination_agent_id: str | None = None,
    request_data_tags: list[str] | None = None,
    capability_token_id: str | None = None,
    capability_action: str = "invoke",
    approval_request_id: str | None = None,
) -> Callable[..., Any]:
    """Wrap an asynchronous Google ADK tool callable."""

    @wraps(fn)
    async def _wrapped(*args: Any, **kwargs: Any) -> Any:
        payload, shape = _normalize_input(args, kwargs)
        tags = list(request_data_tags or [])
        request = self.safeai.intercept_tool_request(
            tool_name=tool_name,
            parameters=payload,
            data_tags=tags,
            agent_id=agent_id,
            session_id=session_id,
            source_agent_id=source_agent_id,
            destination_agent_id=destination_agent_id,
            action_type="google_adk_tool",
            capability_token_id=capability_token_id,
            capability_action=capability_action,
            approval_request_id=approval_request_id,
        )
        if request.decision.action != "allow":
            raise SafeAIBlockedError(
                action=request.decision.action,
                policy_name=request.decision.policy_name,
                reason=request.decision.reason,
            )

        result = await _ainvoke_with_shape(fn, request.filtered_params, shape)
        guarded = self.safeai.intercept_tool_response(
            tool_name=tool_name,
            response=_normalize_response(result),
            agent_id=agent_id,
            request_data_tags=tags,
            session_id=session_id,
            source_agent_id=source_agent_id,
            destination_agent_id=destination_agent_id,
            action_type="google_adk_tool",
        )
        if guarded.decision.action in {"block", "require_approval"}:
            raise SafeAIBlockedError(
                action=guarded.decision.action,
                policy_name=guarded.decision.policy_name,
                reason=guarded.decision.reason,
            )
        return _restore_response_shape(result, guarded.filtered_response)

    return _wrapped

SafeAIBlockedError

Bases: RuntimeError

Raised when SafeAI blocks a LangChain tool invocation.

Source code in safeai/middleware/langchain.py 
21
22
23
24
25
26
27
28
class SafeAIBlockedError(RuntimeError):
    """Raised when SafeAI blocks a LangChain tool invocation."""

    def __init__(self, *, action: str, policy_name: str | None, reason: str) -> None:
        super().__init__(f"SafeAI blocked tool call ({action}): {reason}")
        self.action = action
        self.policy_name = policy_name
        self.reason = reason

SafeAICallback

Bases: BaseCallbackHandler

LangChain callback helper for explicit request/response interception.

Source code in safeai/middleware/langchain.py 
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
class SafeAICallback(BaseCallbackHandler):
    """LangChain callback helper for explicit request/response interception."""

    def __init__(
        self,
        safeai: Any,
        *,
        agent_id: str = "unknown",
        session_id: str | None = None,
        source_agent_id: str | None = None,
        destination_agent_id: str | None = None,
    ) -> None:
        self.safeai = safeai
        self.agent_id = agent_id
        self.session_id = session_id
        self.source_agent_id = source_agent_id
        self.destination_agent_id = destination_agent_id

    def intercept_tool_call(
        self,
        *,
        tool_name: str,
        parameters: dict[str, Any],
        response: dict[str, Any],
        data_tags: list[str],
        capability_token_id: str | None = None,
        capability_action: str = "invoke",
        approval_request_id: str | None = None,
    ) -> dict[str, Any]:
        """Run explicit request + response interception for callback-driven flows."""
        request = self.safeai.intercept_tool_request(
            tool_name=tool_name,
            parameters=parameters,
            data_tags=data_tags,
            agent_id=self.agent_id,
            session_id=self.session_id,
            source_agent_id=self.source_agent_id,
            destination_agent_id=self.destination_agent_id,
            action_type="langchain_callback",
            capability_token_id=capability_token_id,
            capability_action=capability_action,
            approval_request_id=approval_request_id,
        )
        if request.decision.action != "allow":
            raise SafeAIBlockedError(
                action=request.decision.action,
                policy_name=request.decision.policy_name,
                reason=request.decision.reason,
            )

        guarded = self.safeai.intercept_tool_response(
            tool_name=tool_name,
            response=response,
            agent_id=self.agent_id,
            request_data_tags=data_tags,
            session_id=self.session_id,
            source_agent_id=self.source_agent_id,
            destination_agent_id=self.destination_agent_id,
            action_type="langchain_callback",
        )
        if guarded.decision.action in {"block", "require_approval"}:
            raise SafeAIBlockedError(
                action=guarded.decision.action,
                policy_name=guarded.decision.policy_name,
                reason=guarded.decision.reason,
            )
        return guarded.filtered_response

intercept_tool_call 

intercept_tool_call(*, tool_name: str, parameters: dict[str, Any], response: dict[str, Any], data_tags: list[str], capability_token_id: str | None = None, capability_action: str = 'invoke', approval_request_id: str | None = None) -> dict[str, Any]

Run explicit request + response interception for callback-driven flows.

Source code in safeai/middleware/langchain.py 
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
def intercept_tool_call(
    self,
    *,
    tool_name: str,
    parameters: dict[str, Any],
    response: dict[str, Any],
    data_tags: list[str],
    capability_token_id: str | None = None,
    capability_action: str = "invoke",
    approval_request_id: str | None = None,
) -> dict[str, Any]:
    """Run explicit request + response interception for callback-driven flows."""
    request = self.safeai.intercept_tool_request(
        tool_name=tool_name,
        parameters=parameters,
        data_tags=data_tags,
        agent_id=self.agent_id,
        session_id=self.session_id,
        source_agent_id=self.source_agent_id,
        destination_agent_id=self.destination_agent_id,
        action_type="langchain_callback",
        capability_token_id=capability_token_id,
        capability_action=capability_action,
        approval_request_id=approval_request_id,
    )
    if request.decision.action != "allow":
        raise SafeAIBlockedError(
            action=request.decision.action,
            policy_name=request.decision.policy_name,
            reason=request.decision.reason,
        )

    guarded = self.safeai.intercept_tool_response(
        tool_name=tool_name,
        response=response,
        agent_id=self.agent_id,
        request_data_tags=data_tags,
        session_id=self.session_id,
        source_agent_id=self.source_agent_id,
        destination_agent_id=self.destination_agent_id,
        action_type="langchain_callback",
    )
    if guarded.decision.action in {"block", "require_approval"}:
        raise SafeAIBlockedError(
            action=guarded.decision.action,
            policy_name=guarded.decision.policy_name,
            reason=guarded.decision.reason,
        )
    return guarded.filtered_response

SafeAILangChainAdapter

Bases: BaseMiddleware

Adapter that wraps LangChain tool invocations with SafeAI checks.

Source code in safeai/middleware/langchain.py 
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
class SafeAILangChainAdapter(BaseMiddleware):
    """Adapter that wraps LangChain tool invocations with SafeAI checks."""

    def wrap_tool(
        self,
        tool_name: str,
        fn: Callable[..., Any],
        *,
        agent_id: str = "unknown",
        session_id: str | None = None,
        source_agent_id: str | None = None,
        destination_agent_id: str | None = None,
        request_data_tags: list[str] | None = None,
        tag_extractor: TagExtractor | None = None,
        capability_token_id: str | None = None,
        capability_action: str = "invoke",
        approval_request_id: str | None = None,
    ) -> Callable[..., Any]:
        """Wrap a synchronous tool callable.

        Input payload is normalized to key/value form, validated on request,
        then the tool response is filtered before returning to caller.
        """

        @wraps(fn)
        def _wrapped(*args: Any, **kwargs: Any) -> Any:
            payload, shape = _normalize_input(args, kwargs)
            inferred_tags = _infer_tags(payload, safeai=self.safeai, extractor=tag_extractor)
            tags = list(request_data_tags or inferred_tags)

            request = self.safeai.intercept_tool_request(
                tool_name=tool_name,
                parameters=payload,
                data_tags=tags,
                agent_id=agent_id,
                session_id=session_id,
                source_agent_id=source_agent_id,
                destination_agent_id=destination_agent_id,
                action_type="langchain_tool",
                capability_token_id=capability_token_id,
                capability_action=capability_action,
                approval_request_id=approval_request_id,
            )
            if request.decision.action != "allow":
                raise SafeAIBlockedError(
                    action=request.decision.action,
                    policy_name=request.decision.policy_name,
                    reason=request.decision.reason,
                )

            result = _invoke_with_shape(fn, request.filtered_params, shape)
            response_payload = _normalize_response(result)
            guarded = self.safeai.intercept_tool_response(
                tool_name=tool_name,
                response=response_payload,
                agent_id=agent_id,
                request_data_tags=tags,
                session_id=session_id,
                source_agent_id=source_agent_id,
                destination_agent_id=destination_agent_id,
                action_type="langchain_tool",
            )
            if guarded.decision.action in {"block", "require_approval"}:
                raise SafeAIBlockedError(
                    action=guarded.decision.action,
                    policy_name=guarded.decision.policy_name,
                    reason=guarded.decision.reason,
                )
            return _restore_response_shape(result, guarded.filtered_response)

        return _wrapped

    def wrap_async_tool(
        self,
        tool_name: str,
        fn: Callable[..., Any],
        *,
        agent_id: str = "unknown",
        session_id: str | None = None,
        source_agent_id: str | None = None,
        destination_agent_id: str | None = None,
        request_data_tags: list[str] | None = None,
        tag_extractor: TagExtractor | None = None,
        capability_token_id: str | None = None,
        capability_action: str = "invoke",
        approval_request_id: str | None = None,
    ) -> Callable[..., Any]:
        """Wrap an async tool callable."""

        @wraps(fn)
        async def _wrapped(*args: Any, **kwargs: Any) -> Any:
            payload, shape = _normalize_input(args, kwargs)
            inferred_tags = _infer_tags(payload, safeai=self.safeai, extractor=tag_extractor)
            tags = list(request_data_tags or inferred_tags)

            request = self.safeai.intercept_tool_request(
                tool_name=tool_name,
                parameters=payload,
                data_tags=tags,
                agent_id=agent_id,
                session_id=session_id,
                source_agent_id=source_agent_id,
                destination_agent_id=destination_agent_id,
                action_type="langchain_tool",
                capability_token_id=capability_token_id,
                capability_action=capability_action,
                approval_request_id=approval_request_id,
            )
            if request.decision.action != "allow":
                raise SafeAIBlockedError(
                    action=request.decision.action,
                    policy_name=request.decision.policy_name,
                    reason=request.decision.reason,
                )

            result = await _ainvoke_with_shape(fn, request.filtered_params, shape)
            response_payload = _normalize_response(result)
            guarded = self.safeai.intercept_tool_response(
                tool_name=tool_name,
                response=response_payload,
                agent_id=agent_id,
                request_data_tags=tags,
                session_id=session_id,
                source_agent_id=source_agent_id,
                destination_agent_id=destination_agent_id,
                action_type="langchain_tool",
            )
            if guarded.decision.action in {"block", "require_approval"}:
                raise SafeAIBlockedError(
                    action=guarded.decision.action,
                    policy_name=guarded.decision.policy_name,
                    reason=guarded.decision.reason,
                )
            return _restore_response_shape(result, guarded.filtered_response)

        return _wrapped

    def wrap_langchain_tool(
        self,
        tool: Any,
        *,
        agent_id: str = "unknown",
        session_id: str | None = None,
        source_agent_id: str | None = None,
        destination_agent_id: str | None = None,
        request_data_tags: list[str] | None = None,
        capability_token_id: str | None = None,
        capability_action: str = "invoke",
        approval_request_id: str | None = None,
    ) -> Any:
        """Patch `invoke`/`ainvoke` on a LangChain-like tool object in place."""
        tool_name = str(getattr(tool, "name", "") or getattr(tool, "__name__", "tool")).strip() or "tool"
        if hasattr(tool, "invoke") and callable(tool.invoke):
            tool.invoke = self.wrap_tool(
                tool_name=tool_name,
                fn=tool.invoke,
                agent_id=agent_id,
                session_id=session_id,
                source_agent_id=source_agent_id,
                destination_agent_id=destination_agent_id,
                request_data_tags=request_data_tags,
                capability_token_id=capability_token_id,
                capability_action=capability_action,
                approval_request_id=approval_request_id,
            )
        if hasattr(tool, "ainvoke") and callable(tool.ainvoke):
            tool.ainvoke = self.wrap_async_tool(
                tool_name=tool_name,
                fn=tool.ainvoke,
                agent_id=agent_id,
                session_id=session_id,
                source_agent_id=source_agent_id,
                destination_agent_id=destination_agent_id,
                request_data_tags=request_data_tags,
                capability_token_id=capability_token_id,
                capability_action=capability_action,
                approval_request_id=approval_request_id,
            )
        return tool

wrap_tool 

wrap_tool(tool_name: str, fn: Callable[..., Any], *, agent_id: str = 'unknown', session_id: str | None = None, source_agent_id: str | None = None, destination_agent_id: str | None = None, request_data_tags: list[str] | None = None, tag_extractor: TagExtractor | None = None, capability_token_id: str | None = None, capability_action: str = 'invoke', approval_request_id: str | None = None) -> Callable[..., Any]

Wrap a synchronous tool callable.

Input payload is normalized to key/value form, validated on request, then the tool response is filtered before returning to caller.

Source code in safeai/middleware/langchain.py 
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
def wrap_tool(
    self,
    tool_name: str,
    fn: Callable[..., Any],
    *,
    agent_id: str = "unknown",
    session_id: str | None = None,
    source_agent_id: str | None = None,
    destination_agent_id: str | None = None,
    request_data_tags: list[str] | None = None,
    tag_extractor: TagExtractor | None = None,
    capability_token_id: str | None = None,
    capability_action: str = "invoke",
    approval_request_id: str | None = None,
) -> Callable[..., Any]:
    """Wrap a synchronous tool callable.

    Input payload is normalized to key/value form, validated on request,
    then the tool response is filtered before returning to caller.
    """

    @wraps(fn)
    def _wrapped(*args: Any, **kwargs: Any) -> Any:
        payload, shape = _normalize_input(args, kwargs)
        inferred_tags = _infer_tags(payload, safeai=self.safeai, extractor=tag_extractor)
        tags = list(request_data_tags or inferred_tags)

        request = self.safeai.intercept_tool_request(
            tool_name=tool_name,
            parameters=payload,
            data_tags=tags,
            agent_id=agent_id,
            session_id=session_id,
            source_agent_id=source_agent_id,
            destination_agent_id=destination_agent_id,
            action_type="langchain_tool",
            capability_token_id=capability_token_id,
            capability_action=capability_action,
            approval_request_id=approval_request_id,
        )
        if request.decision.action != "allow":
            raise SafeAIBlockedError(
                action=request.decision.action,
                policy_name=request.decision.policy_name,
                reason=request.decision.reason,
            )

        result = _invoke_with_shape(fn, request.filtered_params, shape)
        response_payload = _normalize_response(result)
        guarded = self.safeai.intercept_tool_response(
            tool_name=tool_name,
            response=response_payload,
            agent_id=agent_id,
            request_data_tags=tags,
            session_id=session_id,
            source_agent_id=source_agent_id,
            destination_agent_id=destination_agent_id,
            action_type="langchain_tool",
        )
        if guarded.decision.action in {"block", "require_approval"}:
            raise SafeAIBlockedError(
                action=guarded.decision.action,
                policy_name=guarded.decision.policy_name,
                reason=guarded.decision.reason,
            )
        return _restore_response_shape(result, guarded.filtered_response)

    return _wrapped

wrap_async_tool 

wrap_async_tool(tool_name: str, fn: Callable[..., Any], *, agent_id: str = 'unknown', session_id: str | None = None, source_agent_id: str | None = None, destination_agent_id: str | None = None, request_data_tags: list[str] | None = None, tag_extractor: TagExtractor | None = None, capability_token_id: str | None = None, capability_action: str = 'invoke', approval_request_id: str | None = None) -> Callable[..., Any]

Wrap an async tool callable.

Source code in safeai/middleware/langchain.py 
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
def wrap_async_tool(
    self,
    tool_name: str,
    fn: Callable[..., Any],
    *,
    agent_id: str = "unknown",
    session_id: str | None = None,
    source_agent_id: str | None = None,
    destination_agent_id: str | None = None,
    request_data_tags: list[str] | None = None,
    tag_extractor: TagExtractor | None = None,
    capability_token_id: str | None = None,
    capability_action: str = "invoke",
    approval_request_id: str | None = None,
) -> Callable[..., Any]:
    """Wrap an async tool callable."""

    @wraps(fn)
    async def _wrapped(*args: Any, **kwargs: Any) -> Any:
        payload, shape = _normalize_input(args, kwargs)
        inferred_tags = _infer_tags(payload, safeai=self.safeai, extractor=tag_extractor)
        tags = list(request_data_tags or inferred_tags)

        request = self.safeai.intercept_tool_request(
            tool_name=tool_name,
            parameters=payload,
            data_tags=tags,
            agent_id=agent_id,
            session_id=session_id,
            source_agent_id=source_agent_id,
            destination_agent_id=destination_agent_id,
            action_type="langchain_tool",
            capability_token_id=capability_token_id,
            capability_action=capability_action,
            approval_request_id=approval_request_id,
        )
        if request.decision.action != "allow":
            raise SafeAIBlockedError(
                action=request.decision.action,
                policy_name=request.decision.policy_name,
                reason=request.decision.reason,
            )

        result = await _ainvoke_with_shape(fn, request.filtered_params, shape)
        response_payload = _normalize_response(result)
        guarded = self.safeai.intercept_tool_response(
            tool_name=tool_name,
            response=response_payload,
            agent_id=agent_id,
            request_data_tags=tags,
            session_id=session_id,
            source_agent_id=source_agent_id,
            destination_agent_id=destination_agent_id,
            action_type="langchain_tool",
        )
        if guarded.decision.action in {"block", "require_approval"}:
            raise SafeAIBlockedError(
                action=guarded.decision.action,
                policy_name=guarded.decision.policy_name,
                reason=guarded.decision.reason,
            )
        return _restore_response_shape(result, guarded.filtered_response)

    return _wrapped

wrap_langchain_tool 

wrap_langchain_tool(tool: Any, *, agent_id: str = 'unknown', session_id: str | None = None, source_agent_id: str | None = None, destination_agent_id: str | None = None, request_data_tags: list[str] | None = None, capability_token_id: str | None = None, capability_action: str = 'invoke', approval_request_id: str | None = None) -> Any

Patch invoke/ainvoke on a LangChain-like tool object in place.

Source code in safeai/middleware/langchain.py 
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
def wrap_langchain_tool(
    self,
    tool: Any,
    *,
    agent_id: str = "unknown",
    session_id: str | None = None,
    source_agent_id: str | None = None,
    destination_agent_id: str | None = None,
    request_data_tags: list[str] | None = None,
    capability_token_id: str | None = None,
    capability_action: str = "invoke",
    approval_request_id: str | None = None,
) -> Any:
    """Patch `invoke`/`ainvoke` on a LangChain-like tool object in place."""
    tool_name = str(getattr(tool, "name", "") or getattr(tool, "__name__", "tool")).strip() or "tool"
    if hasattr(tool, "invoke") and callable(tool.invoke):
        tool.invoke = self.wrap_tool(
            tool_name=tool_name,
            fn=tool.invoke,
            agent_id=agent_id,
            session_id=session_id,
            source_agent_id=source_agent_id,
            destination_agent_id=destination_agent_id,
            request_data_tags=request_data_tags,
            capability_token_id=capability_token_id,
            capability_action=capability_action,
            approval_request_id=approval_request_id,
        )
    if hasattr(tool, "ainvoke") and callable(tool.ainvoke):
        tool.ainvoke = self.wrap_async_tool(
            tool_name=tool_name,
            fn=tool.ainvoke,
            agent_id=agent_id,
            session_id=session_id,
            source_agent_id=source_agent_id,
            destination_agent_id=destination_agent_id,
            request_data_tags=request_data_tags,
            capability_token_id=capability_token_id,
            capability_action=capability_action,
            approval_request_id=approval_request_id,
        )
    return tool