Skip to content

naqsha.models

NAP V2 and Thin Model Adapters

naqsha.models

Model Client ports and adapters.

AnthropicMessagesModelClient

Bases: ModelClient

POST {base_url}/v1/messages.

Source code in src/naqsha/models/anthropic.py 
 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
class AnthropicMessagesModelClient(ModelClient):
    """POST ``{base_url}/v1/messages``."""

    def __init__(
        self,
        *,
        base_url: str,
        model: str,
        api_key_env: str = "ANTHROPIC_API_KEY",
        timeout_seconds: float = 120.0,
        max_tokens: int = 4096,
        anthropic_version: str = "2023-06-01",
        post_fn: _PostFn | None = None,
    ) -> None:
        self._base_url = base_url.rstrip("/")
        self._model = model
        self._api_key_env = api_key_env
        self._timeout_seconds = timeout_seconds
        self._max_tokens = max_tokens
        self._anthropic_version = anthropic_version
        self._post_fn = post_fn

    def next_message(
        self,
        *,
        query: str,
        trace: list[TraceEvent],
        tools: list[ToolSpec],
        memory: list[MemoryRecord],
        span_context: SpanContext | None = None,
        instructions: str = "",
    ) -> NapMessage:
        api_key = os.environ.get(self._api_key_env, "").strip()
        if not api_key:
            raise ModelInvocationError(
                f"Environment variable {self._api_key_env!r} is not set or empty."
            )

        transcript = trace_to_transcript(
            query=query, trace=trace, memory=memory, instructions=instructions
        )
        system_text, anthropic_messages = transcript_to_anthropic_messages(transcript)

        url = f"{self._base_url}/v1/messages"
        payload: dict[str, Any] = {
            "model": self._model,
            "max_tokens": self._max_tokens,
            "system": system_text,
            "messages": anthropic_messages,
            "tools": _anthropic_tools(tools),
            "temperature": 0,
        }

        wrapper = self._post_fn if self._post_fn is not None else default_post

        def _post(
            url_: str,
            headers: dict[str, str],
            body: bytes,
            timeout: float,
        ) -> tuple[int, bytes]:
            return wrapper(url_, headers, body, timeout)

        parsed = post_json(
            url,
            headers={
                "x-api-key": api_key,
                "anthropic-version": self._anthropic_version,
            },
            payload=payload,
            timeout_seconds=self._timeout_seconds,
            post_fn=_post,
            error_label="Anthropic Messages API",
        )

        msg_content = parsed.get("content")
        return attach_span_context(_content_blocks_to_nap(msg_content), span_context)

next_message 

next_message(
    *,
    query: str,
    trace: list[TraceEvent],
    tools: list[ToolSpec],
    memory: list[MemoryRecord],
    span_context: SpanContext | None = None,
    instructions: str = "",
) -> NapMessage
Source code in src/naqsha/models/anthropic.py 
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
def next_message(
    self,
    *,
    query: str,
    trace: list[TraceEvent],
    tools: list[ToolSpec],
    memory: list[MemoryRecord],
    span_context: SpanContext | None = None,
    instructions: str = "",
) -> NapMessage:
    api_key = os.environ.get(self._api_key_env, "").strip()
    if not api_key:
        raise ModelInvocationError(
            f"Environment variable {self._api_key_env!r} is not set or empty."
        )

    transcript = trace_to_transcript(
        query=query, trace=trace, memory=memory, instructions=instructions
    )
    system_text, anthropic_messages = transcript_to_anthropic_messages(transcript)

    url = f"{self._base_url}/v1/messages"
    payload: dict[str, Any] = {
        "model": self._model,
        "max_tokens": self._max_tokens,
        "system": system_text,
        "messages": anthropic_messages,
        "tools": _anthropic_tools(tools),
        "temperature": 0,
    }

    wrapper = self._post_fn if self._post_fn is not None else default_post

    def _post(
        url_: str,
        headers: dict[str, str],
        body: bytes,
        timeout: float,
    ) -> tuple[int, bytes]:
        return wrapper(url_, headers, body, timeout)

    parsed = post_json(
        url,
        headers={
            "x-api-key": api_key,
            "anthropic-version": self._anthropic_version,
        },
        payload=payload,
        timeout_seconds=self._timeout_seconds,
        post_fn=_post,
        error_label="Anthropic Messages API",
    )

    msg_content = parsed.get("content")
    return attach_span_context(_content_blocks_to_nap(msg_content), span_context)

ModelInvocationError

Bases: RuntimeError

Raised when transport or provider payloads cannot yield a valid NAP message.

Source code in src/naqsha/models/errors.py 
4
5
class ModelInvocationError(RuntimeError):
    """Raised when transport or provider payloads cannot yield a valid NAP message."""

GeminiGenerateContentModelClient

Bases: ModelClient

POST {base}/v1beta/models/{model}:generateContent.

Source code in src/naqsha/models/gemini.py 
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
class GeminiGenerateContentModelClient(ModelClient):
    """POST ``{base}/v1beta/models/{model}:generateContent``."""

    def __init__(
        self,
        *,
        base_url: str,
        model: str,
        api_key_env: str = "GEMINI_API_KEY",
        timeout_seconds: float = 120.0,
        post_fn: _PostFn | None = None,
    ) -> None:
        self._base_url = base_url.rstrip("/")
        self._model = model
        self._api_key_env = api_key_env
        self._timeout_seconds = timeout_seconds
        self._post_fn = post_fn

    def next_message(
        self,
        *,
        query: str,
        trace: list[TraceEvent],
        tools: list[ToolSpec],
        memory: list[MemoryRecord],
        span_context: SpanContext | None = None,
        instructions: str = "",
    ) -> NapMessage:
        api_key = os.environ.get(self._api_key_env, "").strip()
        if not api_key:
            raise ModelInvocationError(
                f"Environment variable {self._api_key_env!r} is not set or empty."
            )

        transcript = trace_to_transcript(
            query=query, trace=trace, memory=memory, instructions=instructions
        )
        system_instruction, contents = transcript_to_gemini_contents(transcript)

        url = f"{self._base_url}/v1beta/models/{self._model}:generateContent"
        payload: dict[str, Any] = {
            "systemInstruction": system_instruction,
            "contents": contents,
            "tools": [{"functionDeclarations": _gemini_declarations(tools)}],
            "toolConfig": {"functionCallingConfig": {"mode": "AUTO"}},
            "generationConfig": {"temperature": 0},
        }

        wrapper = self._post_fn if self._post_fn is not None else default_post

        def _post(
            url_: str,
            headers: dict[str, str],
            body: bytes,
            timeout: float,
        ) -> tuple[int, bytes]:
            return wrapper(url_, headers, body, timeout)

        parsed = post_json(
            url,
            headers={"x-goog-api-key": api_key},
            payload=payload,
            timeout_seconds=self._timeout_seconds,
            post_fn=_post,
            error_label="Gemini generateContent",
        )

        candidates = parsed.get("candidates")
        if not isinstance(candidates, list) or not candidates:
            raise ModelInvocationError("Gemini response missing candidates[0].")
        cand0 = candidates[0]
        if not isinstance(cand0, dict):
            raise ModelInvocationError("candidates[0] must be an object.")
        inner = cand0.get("content")
        if not isinstance(inner, dict):
            raise ModelInvocationError("candidate.content must be an object.")
        parts = inner.get("parts")
        return attach_span_context(_candidate_parts_to_nap(parts), span_context)

next_message 

next_message(
    *,
    query: str,
    trace: list[TraceEvent],
    tools: list[ToolSpec],
    memory: list[MemoryRecord],
    span_context: SpanContext | None = None,
    instructions: str = "",
) -> NapMessage
Source code in src/naqsha/models/gemini.py 
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
def next_message(
    self,
    *,
    query: str,
    trace: list[TraceEvent],
    tools: list[ToolSpec],
    memory: list[MemoryRecord],
    span_context: SpanContext | None = None,
    instructions: str = "",
) -> NapMessage:
    api_key = os.environ.get(self._api_key_env, "").strip()
    if not api_key:
        raise ModelInvocationError(
            f"Environment variable {self._api_key_env!r} is not set or empty."
        )

    transcript = trace_to_transcript(
        query=query, trace=trace, memory=memory, instructions=instructions
    )
    system_instruction, contents = transcript_to_gemini_contents(transcript)

    url = f"{self._base_url}/v1beta/models/{self._model}:generateContent"
    payload: dict[str, Any] = {
        "systemInstruction": system_instruction,
        "contents": contents,
        "tools": [{"functionDeclarations": _gemini_declarations(tools)}],
        "toolConfig": {"functionCallingConfig": {"mode": "AUTO"}},
        "generationConfig": {"temperature": 0},
    }

    wrapper = self._post_fn if self._post_fn is not None else default_post

    def _post(
        url_: str,
        headers: dict[str, str],
        body: bytes,
        timeout: float,
    ) -> tuple[int, bytes]:
        return wrapper(url_, headers, body, timeout)

    parsed = post_json(
        url,
        headers={"x-goog-api-key": api_key},
        payload=payload,
        timeout_seconds=self._timeout_seconds,
        post_fn=_post,
        error_label="Gemini generateContent",
    )

    candidates = parsed.get("candidates")
    if not isinstance(candidates, list) or not candidates:
        raise ModelInvocationError("Gemini response missing candidates[0].")
    cand0 = candidates[0]
    if not isinstance(cand0, dict):
        raise ModelInvocationError("candidates[0] must be an object.")
    inner = cand0.get("content")
    if not isinstance(inner, dict):
        raise ModelInvocationError("candidate.content must be an object.")
    parts = inner.get("parts")
    return attach_span_context(_candidate_parts_to_nap(parts), span_context)

OllamaChatModelClient

Bases: ModelClient

POST {base_url}/api/chat with tool definitions compatible with Ollama.

Source code in src/naqsha/models/ollama.py 
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
class OllamaChatModelClient(ModelClient):
    """POST ``{base_url}/api/chat`` with tool definitions compatible with Ollama."""

    def __init__(
        self,
        *,
        base_url: str,
        model: str,
        api_key_env: str | None = None,
        timeout_seconds: float = 120.0,
        post_fn: _PostFn | None = None,
    ) -> None:
        self._base_url = base_url.rstrip("/")
        self._model = model
        self._api_key_env = api_key_env
        self._timeout_seconds = timeout_seconds
        self._post_fn = post_fn

    def next_message(
        self,
        *,
        query: str,
        trace: list[TraceEvent],
        tools: list[ToolSpec],
        memory: list[MemoryRecord],
        span_context: SpanContext | None = None,
        instructions: str = "",
    ) -> NapMessage:
        headers: dict[str, str] = {}
        if self._api_key_env:
            api_key = os.environ.get(self._api_key_env, "").strip()
            if not api_key:
                raise ModelInvocationError(
                    f"Environment variable {self._api_key_env!r} is not set or empty."
                )
            headers["Authorization"] = f"Bearer {api_key}"

        messages = trace_to_chat_messages(
            query=query, trace=trace, memory=memory, instructions=instructions
        )
        url = f"{self._base_url}/api/chat"
        payload: dict[str, Any] = {
            "model": self._model,
            "messages": messages,
            "tools": _tools_payload(tools),
            "stream": False,
            "options": {"temperature": 0},
        }
        wrapper = self._post_fn if self._post_fn is not None else default_post

        def _post(
            url_: str,
            hdrs: dict[str, str],
            body: bytes,
            timeout: float,
        ) -> tuple[int, bytes]:
            return wrapper(url_, hdrs, body, timeout)

        parsed = post_json(
            url,
            headers=headers,
            payload=payload,
            timeout_seconds=self._timeout_seconds,
            post_fn=_post,
            error_label="Ollama chat",
        )

        message = parsed.get("message")
        if not isinstance(message, dict):
            raise ModelInvocationError("Ollama response missing message object.")

        return attach_span_context(_openai_message_to_nap(message), span_context)

next_message 

next_message(
    *,
    query: str,
    trace: list[TraceEvent],
    tools: list[ToolSpec],
    memory: list[MemoryRecord],
    span_context: SpanContext | None = None,
    instructions: str = "",
) -> NapMessage
Source code in src/naqsha/models/ollama.py 
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
def next_message(
    self,
    *,
    query: str,
    trace: list[TraceEvent],
    tools: list[ToolSpec],
    memory: list[MemoryRecord],
    span_context: SpanContext | None = None,
    instructions: str = "",
) -> NapMessage:
    headers: dict[str, str] = {}
    if self._api_key_env:
        api_key = os.environ.get(self._api_key_env, "").strip()
        if not api_key:
            raise ModelInvocationError(
                f"Environment variable {self._api_key_env!r} is not set or empty."
            )
        headers["Authorization"] = f"Bearer {api_key}"

    messages = trace_to_chat_messages(
        query=query, trace=trace, memory=memory, instructions=instructions
    )
    url = f"{self._base_url}/api/chat"
    payload: dict[str, Any] = {
        "model": self._model,
        "messages": messages,
        "tools": _tools_payload(tools),
        "stream": False,
        "options": {"temperature": 0},
    }
    wrapper = self._post_fn if self._post_fn is not None else default_post

    def _post(
        url_: str,
        hdrs: dict[str, str],
        body: bytes,
        timeout: float,
    ) -> tuple[int, bytes]:
        return wrapper(url_, hdrs, body, timeout)

    parsed = post_json(
        url,
        headers=headers,
        payload=payload,
        timeout_seconds=self._timeout_seconds,
        post_fn=_post,
        error_label="Ollama chat",
    )

    message = parsed.get("message")
    if not isinstance(message, dict):
        raise ModelInvocationError("Ollama response missing message object.")

    return attach_span_context(_openai_message_to_nap(message), span_context)

OpenAiCompatModelClient

Bases: ModelClient

POST {base_url}/chat/completions on an OpenAI-compatible server.

Source code in src/naqsha/models/openai_compat.py 
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
class OpenAiCompatModelClient(ModelClient):
    """POST ``{base_url}/chat/completions`` on an OpenAI-compatible server."""

    def __init__(
        self,
        *,
        base_url: str,
        model: str,
        api_key_env: str = "OPENAI_API_KEY",
        timeout_seconds: float = 120.0,
        post_fn: _PostFn | None = None,
    ) -> None:
        self._base_url = base_url.rstrip("/")
        self._model = model
        self._api_key_env = api_key_env
        self._timeout_seconds = timeout_seconds
        self._post_fn = post_fn

    def next_message(
        self,
        *,
        query: str,
        trace: list[TraceEvent],
        tools: list[ToolSpec],
        memory: list[MemoryRecord],
        span_context: SpanContext | None = None,
        instructions: str = "",
    ) -> NapMessage:
        api_key = os.environ.get(self._api_key_env, "").strip()
        if not api_key:
            raise ModelInvocationError(
                f"Environment variable {self._api_key_env!r} is not set or empty."
            )

        transcript = trace_to_transcript(
            query=query, trace=trace, memory=memory, instructions=instructions
        )
        messages = transcript_to_openai_chat_messages(transcript)
        url = f"{self._base_url}/chat/completions"
        payload = {
            "model": self._model,
            "messages": messages,
            "tools": _tools_payload(tools),
            "tool_choice": "auto",
            "temperature": 0,
        }
        wrapper = self._post_fn if self._post_fn is not None else default_post

        def _post(
            url_: str,
            headers: dict[str, str],
            body: bytes,
            timeout: float,
        ) -> tuple[int, bytes]:
            return wrapper(url_, headers, body, timeout)

        parsed = post_json(
            url,
            headers={"Authorization": f"Bearer {api_key}"},
            payload=payload,
            timeout_seconds=self._timeout_seconds,
            post_fn=_post,
            error_label="OpenAI-compatible chat completion",
        )

        choices = parsed.get("choices")
        if not isinstance(choices, list) or not choices:
            raise ModelInvocationError("Provider response missing choices[0].")
        first = choices[0]
        if not isinstance(first, dict):
            raise ModelInvocationError("choices[0] must be an object.")
        message = first.get("message")
        if not isinstance(message, dict):
            raise ModelInvocationError("choices[0].message must be an object.")

        return attach_span_context(_openai_message_to_nap(message), span_context)

next_message 

next_message(
    *,
    query: str,
    trace: list[TraceEvent],
    tools: list[ToolSpec],
    memory: list[MemoryRecord],
    span_context: SpanContext | None = None,
    instructions: str = "",
) -> NapMessage
Source code in src/naqsha/models/openai_compat.py 
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
def next_message(
    self,
    *,
    query: str,
    trace: list[TraceEvent],
    tools: list[ToolSpec],
    memory: list[MemoryRecord],
    span_context: SpanContext | None = None,
    instructions: str = "",
) -> NapMessage:
    api_key = os.environ.get(self._api_key_env, "").strip()
    if not api_key:
        raise ModelInvocationError(
            f"Environment variable {self._api_key_env!r} is not set or empty."
        )

    transcript = trace_to_transcript(
        query=query, trace=trace, memory=memory, instructions=instructions
    )
    messages = transcript_to_openai_chat_messages(transcript)
    url = f"{self._base_url}/chat/completions"
    payload = {
        "model": self._model,
        "messages": messages,
        "tools": _tools_payload(tools),
        "tool_choice": "auto",
        "temperature": 0,
    }
    wrapper = self._post_fn if self._post_fn is not None else default_post

    def _post(
        url_: str,
        headers: dict[str, str],
        body: bytes,
        timeout: float,
    ) -> tuple[int, bytes]:
        return wrapper(url_, headers, body, timeout)

    parsed = post_json(
        url,
        headers={"Authorization": f"Bearer {api_key}"},
        payload=payload,
        timeout_seconds=self._timeout_seconds,
        post_fn=_post,
        error_label="OpenAI-compatible chat completion",
    )

    choices = parsed.get("choices")
    if not isinstance(choices, list) or not choices:
        raise ModelInvocationError("Provider response missing choices[0].")
    first = choices[0]
    if not isinstance(first, dict):
        raise ModelInvocationError("choices[0] must be an object.")
    message = first.get("message")
    if not isinstance(message, dict):
        raise ModelInvocationError("choices[0].message must be an object.")

    return attach_span_context(_openai_message_to_nap(message), span_context)