Skip to content

graflo.architecture.query.models

Read-request models for the DB-agnostic graph query surface.

Four questions an agent asks of a live graph — which nodes, what is adjacent to this one, what is reachable from these, how many — expressed once, in logical schema names, and answered identically by every backend.

Filters are :class:FilterExpression, which already renders to AQL, Cypher, nGQL, GSQL, SQL and Python. There is no new query language here, and there is deliberately no way to pass a backend query through: Connection.execute must never be reachable from an agent-facing path, so nothing in this module carries a raw query string.

Validation raises rather than clamps. Silently reducing hops=99 to 3 hands the caller a partial answer it believes is complete — the failure mode this codebase already rejected for TigerGraph edge direction.

Attributes

Classes

AggregateQuery

Bases: GraphQuery

Count or summarise one vertex type.

Source code in graflo/architecture/query/models.py 
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
class AggregateQuery(GraphQuery):
    """Count or summarise one vertex type."""

    vertex_type: str = PydanticField(...)
    function: AggregationType = PydanticField(
        default=AggregationType.COUNT, description="Aggregation to apply."
    )
    group_by: str | None = PydanticField(
        default=None, description="Property to group by. COUNT only."
    )
    aggregated_field: str | None = PydanticField(
        default=None,
        description="Property to aggregate. Required for everything but COUNT.",
    )
    filters: FilterExpression | None = PydanticField(default=None)

    @property
    def aggregation(self) -> AggregationType:
        """`function` as an enum. See `NeighborQuery.edge_direction`."""
        return AggregationType(self.function)

    def _validate_specific(self, caps: QueryCaps) -> None:
        if self.function != AggregationType.COUNT and not self.aggregated_field:
            raise ValueError(
                f"aggregated_field is required for {self.function}; only COUNT "
                "can aggregate without naming a property"
            )
        if self.group_by and self.function != AggregationType.COUNT:
            raise ValueError(
                f"group_by is only supported for COUNT, not {self.function}"
            )

Attributes

aggregated_field = PydanticField(default=None, description='Property to aggregate. Required for everything but COUNT.') class-attribute instance-attribute
aggregation property

function as an enum. See NeighborQuery.edge_direction.

filters = PydanticField(default=None) class-attribute instance-attribute
function = PydanticField(default=AggregationType.COUNT, description='Aggregation to apply.') class-attribute instance-attribute
group_by = PydanticField(default=None, description='Property to group by. COUNT only.') class-attribute instance-attribute
vertex_type = PydanticField(...) class-attribute instance-attribute

GraphQuery

Bases: ConfigBaseModel

Shared envelope for every read request.

Note

There is intentionally no caps field. Caps come from :data:HARD_CAPS and are lowered only through :meth:narrowed, so a request body cannot raise its own ceiling.

Source code in graflo/architecture/query/models.py 
 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
class GraphQuery(ConfigBaseModel):
    """Shared envelope for every read request.

    Note:
        There is intentionally no ``caps`` field. Caps come from
        :data:`HARD_CAPS` and are lowered only through :meth:`narrowed`, so a
        request body cannot raise its own ceiling.
    """

    limit: int = PydanticField(
        default=100, ge=1, description="Rows to return, within `max_rows`."
    )
    timeout_s: float = PydanticField(
        default=10.0, gt=0, description="Seconds to allow, within `timeout_s`."
    )
    projection: list[str] | None = PydanticField(
        default=None,
        description=(
            "Property names to return. None returns whatever the backend "
            "stores, subject to the cap's allow-list."
        ),
    )

    def finish_init(self, caps: QueryCaps | None = None) -> Self:
        """Validate against *caps*, defaulting to the core ceiling.

        Returns:
            Self, so the call composes: ``NodeQuery(...).finish_init()``.

        Raises:
            CapExceededError: Naming the cap that was exceeded.
        """
        effective = caps or HARD_CAPS
        if self.limit > effective.max_rows:
            raise CapExceededError("max_rows", self.limit, effective.max_rows)
        if self.timeout_s > effective.timeout_s:
            raise CapExceededError("timeout_s", self.timeout_s, effective.timeout_s)
        if self.projection is not None and effective.projection_allow_list is not None:
            permitted = set(effective.projection_allow_list)
            denied = sorted(n for n in self.projection if n not in permitted)
            if denied:
                raise CapExceededError(
                    "projection_allow_list", denied, effective.projection_allow_list
                )
        self._validate_specific(effective)
        return self

    def _validate_specific(self, caps: QueryCaps) -> None:
        """Per-query-kind checks. Overridden by subclasses."""

    def narrowed(self, caps: QueryCaps) -> Self:
        """Return a copy fitted to *caps*, raising on anything explicitly asked for.

        The split is what makes caps both enforceable and usable:

        - A value the caller **explicitly set** above a cap raises
          :class:`CapExceededError`. Silently clamping it would hand back a
          partial answer the caller believes is complete — the failure mode this
          codebase already rejected for TigerGraph edge direction.
        - A value the caller **left at its default** is clamped. A policy of
          ``max_rows=5`` must not 422 every request that simply did not mention
          a limit; that would make strict policies unusable rather than safe.

        ``model_fields_set`` is what distinguishes the two, which is precisely
        why it must be consulted rather than comparing against the default.

        Projection is always intersected rather than raising, since an
        allow-list exists to *hide* properties: refusing the request would tell
        the caller which forbidden property they guessed correctly.
        """
        effective = HARD_CAPS.narrow(caps)
        explicit = self.model_fields_set
        narrowed = self.model_copy(deep=True)

        if "limit" in explicit and self.limit > effective.max_rows:
            raise CapExceededError("max_rows", self.limit, effective.max_rows)
        narrowed.limit = min(self.limit, effective.max_rows)

        if "timeout_s" in explicit and self.timeout_s > effective.timeout_s:
            raise CapExceededError("timeout_s", self.timeout_s, effective.timeout_s)
        narrowed.timeout_s = min(self.timeout_s, effective.timeout_s)

        if (
            narrowed.projection is not None
            and effective.projection_allow_list is not None
        ):
            permitted = set(effective.projection_allow_list)
            narrowed.projection = [n for n in narrowed.projection if n in permitted]

        narrowed._narrow_specific(effective, explicit)
        return narrowed

    def _narrow_specific(self, caps: QueryCaps, explicit: set[str]) -> None:
        """Per-query-kind fitting. Overridden by subclasses.

        Args:
            caps: The effective ceiling.
            explicit: Field names the caller actually set, so an explicit
                over-ask can raise while a default is clamped.
        """

Attributes

limit = PydanticField(default=100, ge=1, description='Rows to return, within `max_rows`.') class-attribute instance-attribute
projection = PydanticField(default=None, description="Property names to return. None returns whatever the backend stores, subject to the cap's allow-list.") class-attribute instance-attribute
timeout_s = PydanticField(default=10.0, gt=0, description='Seconds to allow, within `timeout_s`.') class-attribute instance-attribute

Methods:

finish_init(caps=None)

Validate against caps, defaulting to the core ceiling.

Returns:

Type Description
Self

Self, so the call composes: NodeQuery(...).finish_init().

Raises:

Type Description
CapExceededError

Naming the cap that was exceeded.
Source code in graflo/architecture/query/models.py 
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
def finish_init(self, caps: QueryCaps | None = None) -> Self:
    """Validate against *caps*, defaulting to the core ceiling.

    Returns:
        Self, so the call composes: ``NodeQuery(...).finish_init()``.

    Raises:
        CapExceededError: Naming the cap that was exceeded.
    """
    effective = caps or HARD_CAPS
    if self.limit > effective.max_rows:
        raise CapExceededError("max_rows", self.limit, effective.max_rows)
    if self.timeout_s > effective.timeout_s:
        raise CapExceededError("timeout_s", self.timeout_s, effective.timeout_s)
    if self.projection is not None and effective.projection_allow_list is not None:
        permitted = set(effective.projection_allow_list)
        denied = sorted(n for n in self.projection if n not in permitted)
        if denied:
            raise CapExceededError(
                "projection_allow_list", denied, effective.projection_allow_list
            )
    self._validate_specific(effective)
    return self
narrowed(caps)

Return a copy fitted to caps, raising on anything explicitly asked for.

The split is what makes caps both enforceable and usable:

  • A value the caller explicitly set above a cap raises :class:CapExceededError. Silently clamping it would hand back a partial answer the caller believes is complete — the failure mode this codebase already rejected for TigerGraph edge direction.
  • A value the caller left at its default is clamped. A policy of max_rows=5 must not 422 every request that simply did not mention a limit; that would make strict policies unusable rather than safe.

model_fields_set is what distinguishes the two, which is precisely why it must be consulted rather than comparing against the default.

Projection is always intersected rather than raising, since an allow-list exists to hide properties: refusing the request would tell the caller which forbidden property they guessed correctly.

Source code in graflo/architecture/query/models.py 
 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
def narrowed(self, caps: QueryCaps) -> Self:
    """Return a copy fitted to *caps*, raising on anything explicitly asked for.

    The split is what makes caps both enforceable and usable:

    - A value the caller **explicitly set** above a cap raises
      :class:`CapExceededError`. Silently clamping it would hand back a
      partial answer the caller believes is complete — the failure mode this
      codebase already rejected for TigerGraph edge direction.
    - A value the caller **left at its default** is clamped. A policy of
      ``max_rows=5`` must not 422 every request that simply did not mention
      a limit; that would make strict policies unusable rather than safe.

    ``model_fields_set`` is what distinguishes the two, which is precisely
    why it must be consulted rather than comparing against the default.

    Projection is always intersected rather than raising, since an
    allow-list exists to *hide* properties: refusing the request would tell
    the caller which forbidden property they guessed correctly.
    """
    effective = HARD_CAPS.narrow(caps)
    explicit = self.model_fields_set
    narrowed = self.model_copy(deep=True)

    if "limit" in explicit and self.limit > effective.max_rows:
        raise CapExceededError("max_rows", self.limit, effective.max_rows)
    narrowed.limit = min(self.limit, effective.max_rows)

    if "timeout_s" in explicit and self.timeout_s > effective.timeout_s:
        raise CapExceededError("timeout_s", self.timeout_s, effective.timeout_s)
    narrowed.timeout_s = min(self.timeout_s, effective.timeout_s)

    if (
        narrowed.projection is not None
        and effective.projection_allow_list is not None
    ):
        permitted = set(effective.projection_allow_list)
        narrowed.projection = [n for n in narrowed.projection if n in permitted]

    narrowed._narrow_specific(effective, explicit)
    return narrowed

NeighborQuery

Bases: GraphQuery

What is adjacent to one anchor vertex.

The instance-plane counterpart of SchemaGraph.schema_neighbors. They are different questions and must never share a name: this one asks which rows are adjacent, the other which types can be.

Source code in graflo/architecture/query/models.py 
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
class NeighborQuery(GraphQuery):
    """What is adjacent to one anchor vertex.

    The instance-plane counterpart of ``SchemaGraph.schema_neighbors``. They are
    different questions and must never share a name: this one asks which *rows*
    are adjacent, the other which *types* can be.
    """

    vertex_type: str = PydanticField(..., description="Logical type of the anchor.")
    key: str | dict[str, Any] = PydanticField(
        ..., description="Anchor identity value, or a single-field mapping."
    )
    hops: int = PydanticField(
        default=1, ge=1, description="Hop distance, within `max_hops`."
    )
    direction: EdgeDirection = PydanticField(
        default=EdgeDirection.OUT,
        description=(
            "Orientation followed from the anchor. Defaults to OUT, matching "
            "`Connection.fetch_edges`; an edge declared `directed: false` is "
            "followed both ways regardless."
        ),
    )
    edge_relations: list[str] | None = PydanticField(
        default=None, description="Restrict to these relations. None means all."
    )
    filters: FilterExpression | None = PydanticField(default=None)

    @property
    def edge_direction(self) -> EdgeDirection:
        """`direction` as an enum, for handing to a driver.

        `ConfigBaseModel` sets ``use_enum_values=True``, so the stored value is
        the bare string. Backends compare against `EdgeDirection` members, and a
        string silently matches none of them — which each backend then resolves
        differently, so the same query returns different neighbourhoods
        depending on where it runs. Always cross the boundary through this.
        """
        return EdgeDirection(self.direction)

    def _validate_specific(self, caps: QueryCaps) -> None:
        if self.hops > caps.max_hops:
            raise CapExceededError("max_hops", self.hops, caps.max_hops)
        if self.edge_relations and len(self.edge_relations) > caps.max_edge_types:
            raise CapExceededError(
                "max_edge_types", len(self.edge_relations), caps.max_edge_types
            )

    def _narrow_specific(self, caps: QueryCaps, explicit: set[str]) -> None:
        if "hops" in explicit and self.hops > caps.max_hops:
            raise CapExceededError("max_hops", self.hops, caps.max_hops)
        self.hops = min(self.hops, caps.max_hops)
        if self.edge_relations and len(self.edge_relations) > caps.max_edge_types:
            raise CapExceededError(
                "max_edge_types", len(self.edge_relations), caps.max_edge_types
            )

Attributes

direction = PydanticField(default=EdgeDirection.OUT, description='Orientation followed from the anchor. Defaults to OUT, matching `Connection.fetch_edges`; an edge declared `directed: false` is followed both ways regardless.') class-attribute instance-attribute
edge_direction property

direction as an enum, for handing to a driver.

ConfigBaseModel sets use_enum_values=True, so the stored value is the bare string. Backends compare against EdgeDirection members, and a string silently matches none of them — which each backend then resolves differently, so the same query returns different neighbourhoods depending on where it runs. Always cross the boundary through this.

edge_relations = PydanticField(default=None, description='Restrict to these relations. None means all.') class-attribute instance-attribute
filters = PydanticField(default=None) class-attribute instance-attribute
hops = PydanticField(default=1, ge=1, description='Hop distance, within `max_hops`.') class-attribute instance-attribute
key = PydanticField(..., description='Anchor identity value, or a single-field mapping.') class-attribute instance-attribute
vertex_type = PydanticField(..., description='Logical type of the anchor.') class-attribute instance-attribute

NodeQuery

Bases: GraphQuery

List vertices of one type, optionally filtered.

Source code in graflo/architecture/query/models.py 
133
134
135
136
137
138
139
class NodeQuery(GraphQuery):
    """List vertices of one type, optionally filtered."""

    vertex_type: str = PydanticField(..., description="Logical vertex type name.")
    filters: FilterExpression | None = PydanticField(
        default=None, description="Predicate applied before limiting."
    )

Attributes

filters = PydanticField(default=None, description='Predicate applied before limiting.') class-attribute instance-attribute
vertex_type = PydanticField(..., description='Logical vertex type name.') class-attribute instance-attribute

TraverseQuery

Bases: GraphQuery

What is reachable from a set of anchors.

Source code in graflo/architecture/query/models.py 
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
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
class TraverseQuery(GraphQuery):
    """What is reachable from a set of anchors."""

    seeds: list[dict[str, Any]] = PydanticField(
        ...,
        min_length=1,
        description=(
            "Anchors, each `{vertex_type, key}`. Bounded by `max_seeds`: a "
            "traversal fans out per seed, so seed count multiplies cost."
        ),
    )
    max_hops: int = PydanticField(default=2, ge=1)
    direction: EdgeDirection = PydanticField(default=EdgeDirection.ANY)
    edge_relations: list[str] | None = PydanticField(default=None)
    filters: FilterExpression | None = PydanticField(default=None)

    @property
    def edge_direction(self) -> EdgeDirection:
        """`direction` as an enum. See `NeighborQuery.edge_direction`."""
        return EdgeDirection(self.direction)

    def _validate_specific(self, caps: QueryCaps) -> None:
        if self.max_hops > caps.max_hops:
            raise CapExceededError("max_hops", self.max_hops, caps.max_hops)
        if len(self.seeds) > caps.max_seeds:
            raise CapExceededError("max_seeds", len(self.seeds), caps.max_seeds)
        if self.edge_relations and len(self.edge_relations) > caps.max_edge_types:
            raise CapExceededError(
                "max_edge_types", len(self.edge_relations), caps.max_edge_types
            )
        for seed in self.seeds:
            missing = {"vertex_type", "key"} - set(seed)
            if missing:
                raise ValueError(
                    f"seed {seed!r} is missing {sorted(missing)}; each seed needs "
                    "a vertex_type and a key"
                )

    def _narrow_specific(self, caps: QueryCaps, explicit: set[str]) -> None:
        if "max_hops" in explicit and self.max_hops > caps.max_hops:
            raise CapExceededError("max_hops", self.max_hops, caps.max_hops)
        self.max_hops = min(self.max_hops, caps.max_hops)
        # Seeds are always explicit — there is no default set of anchors — so
        # dropping any of them would silently answer a different question.
        if len(self.seeds) > caps.max_seeds:
            raise CapExceededError("max_seeds", len(self.seeds), caps.max_seeds)
        if self.edge_relations and len(self.edge_relations) > caps.max_edge_types:
            raise CapExceededError(
                "max_edge_types", len(self.edge_relations), caps.max_edge_types
            )

Attributes

direction = PydanticField(default=EdgeDirection.ANY) class-attribute instance-attribute
edge_direction property

direction as an enum. See NeighborQuery.edge_direction.

edge_relations = PydanticField(default=None) class-attribute instance-attribute
filters = PydanticField(default=None) class-attribute instance-attribute
max_hops = PydanticField(default=2, ge=1) class-attribute instance-attribute
seeds = PydanticField(..., min_length=1, description='Anchors, each `{vertex_type, key}`. Bounded by `max_seeds`: a traversal fans out per seed, so seed count multiplies cost.') class-attribute instance-attribute