Spaces:
Running
Running
File size: 35,638 Bytes
3193174 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 | """
Building the agent graph from profiles.
Supports:
- Topology validation (cycles, duplicates, directedness)
- Custom node and edge attributes
- Extensible schemas with versioning
- Export to PyG format with arbitrary features
"""
import itertools
from collections.abc import Callable, Sequence
from typing import TYPE_CHECKING, Any
import rustworkx as rx
import torch
from core.schema import (
SCHEMA_VERSION,
AgentNodeSchema,
BaseEdgeSchema,
BaseNodeSchema,
CostMetrics,
EdgeType,
GraphSchema,
LLMConfig,
NodeType,
SchemaValidator,
TaskNodeSchema,
ValidationResult,
WorkflowEdgeSchema,
)
__all__ = [
"BuilderConfig",
"GraphBuilder",
"build_from_adjacency",
"build_from_schema",
"build_property_graph",
"default_edges",
"default_sequence",
]
_TASK_NODE_ID = "__task__"
if TYPE_CHECKING:
from core.graph import RoleGraph
class BuilderConfig:
"""
Graph builder settings.
Controls schema validation, cycle/duplicate tolerance, and feature
widths, as well as default weight behaviour and connection of the
virtual task node.
"""
def __init__(
self,
validate: bool = True,
check_cycles: bool = True,
check_duplicates: bool = True,
allow_self_loops: bool = False,
node_feature_names: list[str] | None = None,
edge_feature_names: list[str] | None = None,
default_edge_dim: int | None = None,
weight_fn: Callable[[str, str, dict], float] | None = None,
default_weight: float = 1.0,
include_task_node: bool = True,
task_edge_weight: float = 1.0,
):
"""
Create the builder configuration.
Args:
validate: Whether to run schema validation before building.
check_cycles: Whether to check for cycles during validation.
check_duplicates: Whether to check for duplicate nodes and edges.
allow_self_loops: Whether to allow self-loops in the graph.
node_feature_names: Node feature names included in the schema.
edge_feature_names: Edge feature names included in the schema.
default_edge_dim: Default dimensionality of edge feature vectors.
weight_fn: Custom function for computing edge weights.
default_weight: Edge weight when weight_fn is not set.
include_task_node: Whether to add a virtual task node.
task_edge_weight: Weight of edges between the task and agents.
"""
self.validate = validate
self.check_cycles = check_cycles
self.check_duplicates = check_duplicates
self.allow_self_loops = allow_self_loops
self.node_feature_names = node_feature_names or []
self.edge_feature_names = edge_feature_names or []
self.default_edge_dim = default_edge_dim
self.weight_fn = weight_fn
self.default_weight = default_weight
self.include_task_node = include_task_node
self.task_edge_weight = task_edge_weight
class GraphBuilder:
"""
Convenient interface for step-by-step construction of `GraphSchema`.
Supports conditional routing via
add_conditional_edge and add_conditional_edges methods.
Supports explicit start/end nodes for execution optimisation.
Example:
builder = GraphBuilder()
builder.add_agent("solver", description="Solves problems")
builder.add_agent("reviewer", description="Reviews solutions")
builder.add_agent("finalizer", description="Finalizes answer")
# Unconditional edges
builder.add_workflow_edge("solver", "reviewer")
# Conditional edge: transition to finalizer only if reviewer succeeds
builder.add_conditional_edge(
"reviewer", "finalizer",
condition=lambda ctx: "approved" in ctx.get_last_response().lower()
)
# Set explicit execution boundaries
builder.set_start_node("solver")
builder.set_end_node("finalizer")
graph = builder.build()
"""
def __init__(self, config: BuilderConfig | None = None):
"""Initialise the builder with the given configuration."""
self.config = config or BuilderConfig()
self._schema = GraphSchema(
schema_version=SCHEMA_VERSION,
node_feature_names=self.config.node_feature_names,
edge_feature_names=self.config.edge_feature_names,
)
self._validator = SchemaValidator(
check_cycles=self.config.check_cycles,
check_duplicates=self.config.check_duplicates,
)
# Callable conditions are stored separately (not serialised in the schema)
self._edge_conditions: dict[tuple[str, str], Callable] = {}
# Explicit graph execution boundaries
self._start_node: str | None = None
self._end_node: str | None = None
# Agent profiles (for passing tool objects)
self._agent_profiles: dict[str, Any] = {}
def add_agent(
self,
agent_id: str,
display_name: str | None = None,
persona: str = "",
description: str = "",
embedding: list[float] | None = None,
trust_score: float = 1.0,
llm_config: LLMConfig | None = None,
tools: list[str] | None = None,
input_schema: Any | None = None,
output_schema: Any | None = None,
llm_backbone: str | None = None,
base_url: str | None = None,
api_key: str | None = None,
max_tokens: int | None = None,
temperature: float | None = None,
timeout: float | None = None,
top_p: float | None = None,
stop_sequences: list[str] | None = None,
**metadata,
) -> "GraphBuilder":
"""
Add an agent node to the schema with optional LLM configuration and validation.
Args:
agent_id: Unique agent identifier.
display_name: Display name (defaults to the id).
persona: Brief persona/role.
description: Text description of capabilities.
embedding: Agent feature vector (if already computed).
trust_score: Base trust score for the agent.
llm_config: Ready-made LLMConfig object (overrides individual params).
tools: List of agent tools.
input_schema: Pydantic model or JSON Schema for validating incoming data.
output_schema: Pydantic model or JSON Schema for validating agent responses.
llm_backbone: Model name (e.g. "gpt-4", "claude-3-opus").
base_url: API endpoint URL (e.g. "https://api.openai.com/v1").
api_key: API key or reference to an environment variable ("$OPENAI_API_KEY").
max_tokens: Maximum number of tokens in the response.
temperature: Generation temperature (0.0-2.0).
timeout: Request timeout in seconds.
top_p: Top-p (nucleus) sampling parameter.
stop_sequences: Sequences to stop generation.
**metadata: Arbitrary additional fields.
Example:
# Agent with OpenAI GPT-4
builder.add_agent(
"solver",
persona="Expert problem solver",
llm_backbone="gpt-4",
base_url="https://api.openai.com/v1",
api_key="$OPENAI_API_KEY",
temperature=0.7,
max_tokens=2000
)
# Agent with a local Ollama model
builder.add_agent(
"analyzer",
persona="Data analyzer",
llm_backbone="llama3:70b",
base_url="http://localhost:11434/v1",
temperature=0.0
)
# Agent with a ready-made LLMConfig
config = LLMConfig(model_name="claude-3-opus", ...)
builder.add_agent("reviewer", llm_config=config)
# Agent with input/output validation
from pydantic import BaseModel
class SolverInput(BaseModel):
question: str
context: str | None = None
class SolverOutput(BaseModel):
answer: str
confidence: float
builder.add_agent(
"solver",
persona="Math solver",
input_schema=SolverInput,
output_schema=SolverOutput
)
"""
# If llm_config provided, use it as fallback
effective_llm_backbone = llm_backbone
effective_base_url = base_url
effective_api_key = api_key
effective_max_tokens = max_tokens
effective_temperature = temperature
effective_timeout = timeout
effective_top_p = top_p
effective_stop_sequences = stop_sequences
if llm_config:
effective_llm_backbone = llm_backbone or llm_config.model_name
effective_base_url = base_url or llm_config.base_url
effective_api_key = api_key or llm_config.api_key
effective_max_tokens = max_tokens if max_tokens is not None else llm_config.max_tokens
effective_temperature = temperature if temperature is not None else llm_config.temperature
effective_timeout = timeout if timeout is not None else llm_config.timeout
effective_top_p = top_p if top_p is not None else llm_config.top_p
effective_stop_sequences = stop_sequences or llm_config.stop_sequences
node = AgentNodeSchema(
id=agent_id,
display_name=display_name or agent_id,
persona=persona,
description=description,
embedding=embedding,
trust_score=trust_score,
llm_backbone=effective_llm_backbone,
base_url=effective_base_url,
api_key=effective_api_key,
max_tokens=effective_max_tokens,
temperature=effective_temperature,
timeout=effective_timeout,
top_p=effective_top_p,
stop_sequences=effective_stop_sequences,
tools=tools or [],
input_schema=input_schema,
output_schema=output_schema,
metadata=metadata,
)
self._schema.add_node(node)
return self
def add_agent_profile(
self,
profile: Any, # AgentProfile
trust_score: float = 1.0,
**metadata,
) -> "GraphBuilder":
"""
Add an agent from an existing AgentProfile object.
Args:
profile: AgentProfile object.
trust_score: Base trust score for the agent.
**metadata: Additional fields.
Example:
from core.agent import AgentProfile
from tools import CodeInterpreterTool
agent = AgentProfile(
agent_id="coder",
display_name="Coder Agent",
persona="a Python programmer",
tools=[CodeInterpreterTool()],
)
builder = GraphBuilder()
builder.add_agent_profile(agent)
"""
# Get tool names (strings or names from objects)
tool_names = []
if hasattr(profile, "get_tool_names"):
tool_names = profile.get_tool_names()
elif hasattr(profile, "tools"):
tool_names = profile.tools
# Get LLM config if present
llm_config = None
if hasattr(profile, "llm_config") and profile.llm_config:
llm_config = profile.llm_config
node = AgentNodeSchema(
id=profile.agent_id,
display_name=profile.display_name,
persona=getattr(profile, "persona", ""),
description=getattr(profile, "description", ""),
embedding=None, # Will be computed later if needed
trust_score=trust_score,
llm_backbone=getattr(profile, "llm_backbone", None),
base_url=llm_config.base_url if llm_config else None,
api_key=llm_config.api_key if llm_config else None,
max_tokens=llm_config.max_tokens if llm_config else None,
temperature=llm_config.temperature if llm_config else None,
timeout=llm_config.timeout if llm_config else None,
top_p=llm_config.top_p if llm_config else None,
stop_sequences=llm_config.stop_sequences if llm_config else None,
tools=tool_names,
input_schema=getattr(profile, "input_schema", None),
output_schema=getattr(profile, "output_schema", None),
metadata=metadata,
)
self._schema.add_node(node)
# Save the original profile for passing tool objects to the runner
self._agent_profiles[profile.agent_id] = profile
return self
def add_task(
self,
task_id: str = _TASK_NODE_ID,
query: str = "",
description: str = "",
embedding: list[float] | None = None,
**metadata,
) -> "GraphBuilder":
"""
Add a virtual task node.
Args:
task_id: Task node identifier.
query: Task query/formulation.
description: Additional context description.
embedding: Task feature vector.
**metadata: Additional arbitrary fields.
"""
node = TaskNodeSchema(
id=task_id,
query=query,
description=description,
embedding=embedding,
metadata=metadata,
)
self._schema.add_node(node)
return self
def add_node(
self,
node_id: str,
node_type: NodeType = NodeType.CUSTOM,
**kwargs,
) -> "GraphBuilder":
"""Add an arbitrary node of the specified type."""
node = BaseNodeSchema(
id=node_id,
type=node_type,
**kwargs,
)
self._schema.add_node(node)
return self
def add_edge(
self,
source: str,
target: str,
edge_type: EdgeType = EdgeType.WORKFLOW,
weight: float | None = None,
probability: float = 1.0,
cost: CostMetrics | None = None,
attr: list[float] | None = None,
**metadata,
) -> "GraphBuilder":
"""
Add a basic edge between nodes.
Args:
source: Edge source.
target: Edge target.
edge_type: Edge type (workflow/task/...).
weight: Edge weight; if None, computed by weight_fn or default.
probability: Transmission/activation probability.
cost: Transition cost metrics.
attr: Edge feature vector.
**metadata: Additional metadata written to the schema.
"""
if weight is None:
if self.config.weight_fn:
weight = self.config.weight_fn(source, target, metadata)
else:
weight = self.config.default_weight
if source == target and not self.config.allow_self_loops:
msg = f"Self-loops not allowed: {source} -> {target}"
raise ValueError(msg)
edge = BaseEdgeSchema(
source=source,
target=target,
type=edge_type,
weight=weight,
probability=probability,
cost=cost or CostMetrics(),
attr=attr,
metadata=metadata,
)
self._schema.add_edge(edge)
return self
def add_workflow_edge(
self,
source: str,
target: str,
weight: float | None = None,
condition: str | None = None,
priority: int = 0,
**metadata,
) -> "GraphBuilder":
"""
Add a workflow (business logic) edge.
Args:
source: Source node.
target: Target node.
weight: Weight; if None β computed as for a regular edge.
condition: Activation condition (e.g. a route filter).
priority: Execution priority within the workflow.
**metadata: Additional schema attributes.
"""
if weight is None:
if self.config.weight_fn:
weight = self.config.weight_fn(source, target, metadata)
else:
weight = self.config.default_weight
edge = WorkflowEdgeSchema(
source=source,
target=target,
weight=weight,
condition=condition,
priority=priority,
metadata=metadata,
)
self._schema.add_edge(edge)
return self
def add_conditional_edge(
self,
source: str,
target: str,
condition: Callable | str,
weight: float | None = None,
priority: int = 0,
**metadata,
) -> "GraphBuilder":
"""
Add a conditional edge.
The condition is evaluated at runtime during execution plan building.
If the condition is not met, the edge is ignored.
Args:
source: Source node.
target: Target node.
condition: Activation condition:
- Callable[[ConditionContext], bool] β check function
- str β name of a registered condition or expression
weight: Edge weight (if None β default_weight).
priority: Edge priority.
**metadata: Additional attributes.
Returns:
self for chaining.
Example:
# Callable condition
builder.add_conditional_edge(
"solver", "reviewer",
condition=lambda ctx: ctx.source_succeeded()
)
# String condition (built-in)
builder.add_conditional_edge(
"reviewer", "finalizer",
condition="source_success"
)
# String condition (content check)
builder.add_conditional_edge(
"classifier", "math_agent",
condition="contains:math"
)
"""
if weight is None:
if self.config.weight_fn:
weight = self.config.weight_fn(source, target, metadata)
else:
weight = self.config.default_weight
# String condition saved in the schema
condition_str = condition if isinstance(condition, str) else None
edge = WorkflowEdgeSchema(
source=source,
target=target,
weight=weight,
condition=condition_str,
priority=priority,
is_conditional=True,
metadata=metadata,
)
self._schema.add_edge(edge)
# Callable condition saved separately
if callable(condition):
self._edge_conditions[(source, target)] = condition
return self
def add_conditional_edges(
self,
source: str,
path_map: dict[str, Callable | str | None],
default: str | None = None,
weight: float | None = None,
) -> "GraphBuilder":
"""
Add multiple conditional edges from one source (router pattern).
Analogous to LangGraph add_conditional_edges β selecting the next node
based on conditions.
Args:
source: Source node.
path_map: Dict {target: condition}.
condition=None means unconditional transition.
default: Default node if no condition is met.
weight: Weight for all edges.
Returns:
self for chaining.
Example:
# Router: classifier routes to different agents
builder.add_conditional_edges(
"classifier",
{
"math_agent": lambda ctx: "math" in ctx.get_last_response(),
"code_agent": lambda ctx: "code" in ctx.get_last_response(),
"general_agent": None, # fallback
},
default="general_agent"
)
"""
for target, condition in path_map.items():
if condition is None:
# Unconditional edge
self.add_workflow_edge(source, target, weight=weight)
else:
self.add_conditional_edge(source, target, condition, weight=weight)
# If there is a default and it has not been added yet
if default and default not in path_map:
self.add_workflow_edge(source, default, weight=weight)
return self
@property
def edge_conditions(self) -> dict[tuple[str, str], Callable]:
"""Get all callable conditions for edges."""
return self._edge_conditions.copy()
def set_start_node(self, node_id: str) -> "GraphBuilder":
"""
Set the start node for execution.
The start node is the graph entry point. Nodes unreachable from
the start node will be excluded from execution.
Args:
node_id: Node ID (must be added before calling build()).
Returns:
self for chaining.
Example:
builder.add_agent("input_agent", ...)
builder.set_start_node("input_agent")
"""
self._start_node = node_id
return self
def set_end_node(self, node_id: str) -> "GraphBuilder":
"""
Set the end node for execution.
The end node is the graph exit point. Nodes from which
the end node is unreachable will be excluded from execution.
Args:
node_id: Node ID (must be added before calling build()).
Returns:
self for chaining.
Example:
builder.add_agent("output_agent", ...)
builder.set_end_node("output_agent")
"""
self._end_node = node_id
return self
def set_execution_bounds(
self,
start_node: str | None,
end_node: str | None,
) -> "GraphBuilder":
"""
Set execution boundaries (start and end nodes).
Args:
start_node: Start node ID (None for auto-detection).
end_node: End node ID (None for auto-detection).
Returns:
self for chaining.
Example:
builder.set_execution_bounds("input", "output")
"""
self._start_node = start_node
self._end_node = end_node
return self
@property
def start_node(self) -> str | None:
"""Get the start node ID."""
return self._start_node
@property
def end_node(self) -> str | None:
"""Get the end node ID."""
return self._end_node
def connect_task_to_agents(
self,
task_id: str = _TASK_NODE_ID,
agent_ids: list[str] | None = None,
bidirectional: bool = True,
) -> "GraphBuilder":
"""
Connect the task node to agents via context/update edges.
Args:
task_id: Task node identifier.
agent_ids: List of agents; if None, all schema agents are used.
bidirectional: Whether to add reverse edges from agents to the task.
"""
if agent_ids is None:
agent_ids = [node_id for node_id, node in self._schema.nodes.items() if node.type == NodeType.AGENT]
for agent_id in agent_ids:
self.add_edge(
task_id,
agent_id,
edge_type=EdgeType.TASK_CONTEXT,
weight=self.config.task_edge_weight,
)
if bidirectional:
self.add_edge(
agent_id,
task_id,
edge_type=EdgeType.TASK_UPDATE,
weight=self.config.task_edge_weight,
)
return self
def from_edges(
self,
edges: Sequence[tuple[str, str]],
weight: float | None = None,
) -> "GraphBuilder":
"""Add a set of workflow edges from a list of (source, target) pairs."""
for source, target in edges:
self.add_workflow_edge(source, target, weight=weight)
return self
def validate(self) -> ValidationResult:
"""Validate the schema for correctness and return the validation result."""
return self._validator.validate(self._schema)
def build(self) -> "RoleGraph":
"""Build a `RoleGraph` from the current schema, optionally validating it."""
if self.config.validate:
result = self.validate()
if not result.valid:
msg = f"Schema validation failed: {result.errors}"
raise ValueError(msg)
return build_from_schema(
self._schema,
edge_conditions=self._edge_conditions if self._edge_conditions else None,
start_node=self._start_node,
end_node=self._end_node,
)
@property
def schema(self) -> GraphSchema:
"""Current graph schema (for reading/extending outside the builder)."""
return self._schema
def build_from_schema(
schema: GraphSchema,
edge_conditions: dict[tuple[str, str], Callable] | None = None,
start_node: str | None = None,
end_node: str | None = None,
) -> "RoleGraph":
"""
Construct a `RoleGraph` from a ready-made `GraphSchema`.
Creates an internal `rx.PyDiGraph`, transfers node/edge data,
restores adjacency and probability matrices, and builds
a `RoleGraph` with the corresponding agent and task objects.
LLM configuration from AgentNodeSchema is automatically transferred
to AgentProfile.llm_config to support multi-model usage.
Args:
schema: Graph schema.
edge_conditions: Callable conditions for edges.
start_node: Start node ID for execution.
end_node: End node ID for execution.
"""
from core.agent import AgentLLMConfig, AgentProfile, TaskNode
from core.graph import RoleGraph
graph = rx.PyDiGraph()
idx_map = {}
agents = []
task_node = None
task_node_id = None
for node_id, node_schema in schema.nodes.items():
idx_map[node_id] = graph.add_node(
{
"id": node_id,
"type": node_schema.type.value,
"schema": node_schema.model_dump(),
}
)
embedding = None
if node_schema.embedding:
embedding = torch.tensor(node_schema.embedding, dtype=torch.float32)
if node_schema.type == NodeType.TASK:
task_schema = node_schema
task_node = TaskNode(
agent_id=node_id,
query=getattr(task_schema, "query", ""),
description=getattr(task_schema, "description", ""),
embedding=embedding,
)
task_node_id = node_id
agents.append(task_node)
else:
agent_schema = node_schema
# Build LLM config from schema if any LLM params are set
llm_config = None
if isinstance(agent_schema, AgentNodeSchema) and agent_schema.has_llm_config():
llm_config = AgentLLMConfig(
model_name=agent_schema.llm_backbone,
base_url=agent_schema.base_url,
api_key=agent_schema.api_key,
max_tokens=agent_schema.max_tokens,
temperature=agent_schema.temperature,
timeout=agent_schema.timeout,
top_p=agent_schema.top_p,
stop_sequences=agent_schema.stop_sequences,
)
agent = AgentProfile(
agent_id=node_id,
display_name=node_schema.display_name or node_id,
persona=getattr(agent_schema, "persona", ""),
description=getattr(agent_schema, "description", ""),
llm_backbone=getattr(agent_schema, "llm_backbone", None),
llm_config=llm_config,
tools=getattr(agent_schema, "tools", []),
embedding=embedding,
input_schema=getattr(agent_schema, "input_schema", None),
output_schema=getattr(agent_schema, "output_schema", None),
)
agents.append(agent)
for edge_schema in schema.edges:
if edge_schema.source in idx_map and edge_schema.target in idx_map:
edge_data = {
"type": edge_schema.type.value,
"weight": edge_schema.weight,
"probability": edge_schema.probability,
"attr": edge_schema.to_attr_tensor(),
"schema": edge_schema.model_dump(exclude={"attr"}),
}
graph.add_edge(
idx_map[edge_schema.source],
idx_map[edge_schema.target],
edge_data,
)
n = graph.num_nodes()
a_com = torch.zeros((n, n), dtype=torch.float32)
p_matrix = torch.zeros((n, n), dtype=torch.float32)
for edge_idx in graph.edge_indices():
s, t = graph.get_edge_endpoints_by_index(edge_idx)
d = graph.get_edge_data_by_index(edge_idx)
a_com[s, t] = d.get("weight", 1.0)
p_matrix[s, t] = d.get("probability", 1.0)
connections = {a.agent_id: [] for a in agents}
edge_condition_names: dict[tuple[str, str], str] = {}
for edge_schema in schema.edges:
if edge_schema.source in connections and edge_schema.target not in connections[edge_schema.source]:
connections[edge_schema.source].append(edge_schema.target)
# Extract string conditions from WorkflowEdgeSchema
if hasattr(edge_schema, "condition") and edge_schema.condition and isinstance(edge_schema.condition, str):
edge_condition_names[(edge_schema.source, edge_schema.target)] = edge_schema.condition
return RoleGraph(
agents=agents,
node_ids=[a.agent_id for a in agents],
role_connections=connections,
task_node=task_node_id,
query=getattr(task_node, "query", "") if task_node else "",
graph=graph,
A_com=a_com,
p_matrix=p_matrix,
edge_conditions=edge_conditions or {},
edge_condition_names=edge_condition_names,
start_node=start_node,
end_node=end_node,
)
def default_sequence(roles: Sequence[str], anchor: str) -> list[str]:
"""Build the role order starting from the anchor (if present)."""
ordered = [anchor] if anchor in roles else []
ordered.extend(role for role in roles if role != anchor)
return ordered
def default_edges(sequence: Sequence[str]) -> list[tuple[str, str]]:
"""Build a chain of edges from a sequence of nodes (s -> t in order)."""
return list(itertools.pairwise(sequence))
def build_property_graph(
agents: Sequence[Any],
workflow_edges: Sequence[tuple[str, str]],
*,
query: str = "",
answer: str = "",
anchor: str | None = None,
encoder: Any | None = None,
include_task_node: bool = True,
config: BuilderConfig | None = None,
node_features: dict[str, dict[str, Any]] | None = None,
edge_features: dict[tuple[str, str], dict[str, Any]] | None = None,
) -> "RoleGraph":
"""
Build a `RoleGraph` with the given workflow topology.
Args:
agents: List of agents (`AgentProfile`).
workflow_edges: Workflow edges [(source, target), ...].
query: Task text placed in the task node.
answer: Known answer (added to the final graph).
anchor: Anchor agent ID β will be first in order and may boost edge weights.
encoder: `NodeEncoder` for automatic embedding computation.
include_task_node: Whether to add a virtual task node and connect it to agents.
config: Builder configuration (if not provided β a new one is created).
node_features: Additional node features `{node_id: {feature: value}}`.
edge_features: Additional edge features `{(src, tgt): {feature: value}}`.
Returns:
Ready-made `RoleGraph` with populated adjacency and probability matrices.
"""
config = config or BuilderConfig(include_task_node=include_task_node)
builder = GraphBuilder(config)
if encoder is not None:
from core.agent import TaskNode
texts = [a.to_text() for a in agents]
if include_task_node:
task_tmpl = TaskNode(query=query)
texts.append(task_tmpl.to_text())
embs = encoder.encode(texts)
emb_list = [embs[i].cpu().tolist() for i in range(len(agents))]
task_emb = embs[-1].cpu().tolist() if include_task_node else None
else:
emb_list = [None] * len(agents)
task_emb = None
if anchor is None and agents:
anchor = agents[0].agent_id
for i, agent in enumerate(agents):
extra = (node_features or {}).get(agent.agent_id, {})
trust = 1.0 + (0.1 if agent.agent_id == anchor else 0)
builder.add_agent(
agent_id=agent.agent_id,
display_name=agent.display_name,
persona=getattr(agent, "persona", ""),
description=getattr(agent, "description", ""),
embedding=emb_list[i],
trust_score=min(trust, 1.0),
llm_backbone=getattr(agent, "llm_backbone", None),
tools=getattr(agent, "tools", []),
**extra,
)
if include_task_node:
task_extra = (node_features or {}).get(_TASK_NODE_ID, {})
builder.add_task(
task_id=_TASK_NODE_ID,
query=query,
embedding=task_emb,
**task_extra,
)
pos = {a.agent_id: i for i, a in enumerate(agents)}
for s, t in workflow_edges:
if s not in pos or t not in pos:
continue
w = 1.0
if anchor and s == anchor:
w += 0.1
if pos.get(s, 0) <= pos.get(t, 0):
w += 0.05
extra = (edge_features or {}).get((s, t), {})
builder.add_workflow_edge(
source=s,
target=t,
weight=round(w, 3),
**extra,
)
if include_task_node:
builder.connect_task_to_agents(_TASK_NODE_ID, bidirectional=True)
rg = builder.build()
if answer:
object.__setattr__(rg, "answer", answer)
return rg
def build_from_adjacency(
agents: Sequence[Any],
adjacency: torch.Tensor,
*,
query: str = "",
answer: str = "",
threshold: float = 0.5,
config: BuilderConfig | None = None,
) -> "RoleGraph":
"""
Build a `RoleGraph` from an adjacency matrix.
Args:
agents: List of agents (`AgentProfile`).
adjacency: Square weight/connectivity matrix (torch.Tensor).
query: Task text placed in the task node.
answer: Known answer (added to the final graph).
threshold: Threshold above which a connection is considered to exist.
config: Builder configuration (task node is disabled).
"""
n = len(agents)
if tuple(adjacency.shape) != (n, n):
msg = f"Adjacency shape {tuple(adjacency.shape)} doesn't match {n} agents"
raise ValueError(msg)
edges = [
(agents[i].agent_id, agents[j].agent_id)
for i in range(n)
for j in range(n)
if i != j and adjacency[i, j] > threshold
]
config = config or BuilderConfig(include_task_node=False)
config.include_task_node = False
return build_property_graph(
agents,
edges,
query=query,
answer=answer,
config=config,
)
|