first
This commit is contained in:
valenti
@@ -0,0 +1,21 @@
|
||||
"""asf-context: Context Engine — "l'impiegato che raccoglie il fascicolo completo".
|
||||
|
||||
Given an artifact / project / user question, assembles relevant context for
|
||||
AI operations. In the MVP this is pure MongoDB relationship traversal.
|
||||
|
||||
The vector / semantic retrieval logic lives behind ``SemanticStore`` protocol
|
||||
so a vector DB can be plugged in later without changing callers.
|
||||
"""
|
||||
|
||||
__version__ = "0.1.0"
|
||||
|
||||
from .context_engine import ContextEngine, AssemblyOptions, ContextAssembly
|
||||
from .semantic_store import SemanticStore, InMemorySemanticStoreMock
|
||||
|
||||
__all__ = [
|
||||
"ContextEngine",
|
||||
"AssemblyOptions",
|
||||
"ContextAssembly",
|
||||
"SemanticStore",
|
||||
"InMemorySemanticStoreMock",
|
||||
]
|
||||
@@ -0,0 +1,28 @@
|
||||
"""Minimal configuration surface for asf-context service.
|
||||
|
||||
Keeps parity with asf-core config. Can be grown independently.
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
from functools import lru_cache
|
||||
from pydantic_settings import BaseSettings, SettingsConfigDict
|
||||
|
||||
|
||||
class ContextSettings(BaseSettings):
|
||||
mongo_uri: str = "mongodb://172.19.40.102:27017,172.19.40.103:27017/asf?replicaSet=rs0"
|
||||
mongo_db: str = "asf"
|
||||
|
||||
rabbitmq_uri: str = "amqp://azurian:StrongPassword123!@172.19.40.201:5672/asf"
|
||||
rabbitmq_events_exchange: str = "asf.events"
|
||||
|
||||
model_config = SettingsConfigDict(
|
||||
env_file=".env",
|
||||
env_file_encoding="utf-8",
|
||||
extra="ignore",
|
||||
)
|
||||
|
||||
|
||||
@lru_cache(maxsize=1)
|
||||
def get_settings() -> ContextSettings:
|
||||
return ContextSettings()
|
||||
@@ -0,0 +1,236 @@
|
||||
"""ContextEngine: assembles "il fascicolo completo" for an AI operation.
|
||||
|
||||
Input: an artifact (or project id, or a free-text question)
|
||||
Output: a ``ContextAssembly`` bundling
|
||||
· related requirements
|
||||
· architecture decisions
|
||||
· relevant source files
|
||||
· docs
|
||||
· tests
|
||||
· deployments
|
||||
· previous AI analysis
|
||||
· incidents
|
||||
· runtime info (currently placeholder)
|
||||
|
||||
For MVP it walks the ArtifactRelationship graph in MongoDB. Semantic
|
||||
retrieval is hidden behind ``SemanticStore`` so we can add embeddings later
|
||||
without changing the engine surface.
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
from dataclasses import dataclass, field
|
||||
from typing import Optional, Sequence
|
||||
from uuid import UUID
|
||||
|
||||
import structlog
|
||||
|
||||
from asf_domain import Artifact, ArtifactRelationship, RelationshipType
|
||||
from .semantic_store import SemanticStore, InMemorySemanticStoreMock, SemanticHit
|
||||
|
||||
log = structlog.get_logger(__name__)
|
||||
|
||||
|
||||
@dataclass
|
||||
class AssemblyOptions:
|
||||
max_depth: int = 2
|
||||
max_nodes: int = 200
|
||||
include_semantic: bool = False
|
||||
semantic_top_k: int = 10
|
||||
relationship_whitelist: Optional[Sequence[RelationshipType]] = None
|
||||
|
||||
|
||||
@dataclass
|
||||
class ContextAssembly:
|
||||
seed_artifact_id: Optional[UUID]
|
||||
project_id: Optional[UUID]
|
||||
tenant_id: UUID
|
||||
|
||||
# Artifacts grouped by logical concern (all are Artifact objects)
|
||||
requirements: list[Artifact] = field(default_factory=list)
|
||||
architecture_decisions: list[Artifact] = field(default_factory=list)
|
||||
tasks: list[Artifact] = field(default_factory=list)
|
||||
source_files: list[Artifact] = field(default_factory=list)
|
||||
documents: list[Artifact] = field(default_factory=list)
|
||||
tests: list[Artifact] = field(default_factory=list)
|
||||
builds: list[Artifact] = field(default_factory=list)
|
||||
deployments: list[Artifact] = field(default_factory=list)
|
||||
releases: list[Artifact] = field(default_factory=list)
|
||||
incidents: list[Artifact] = field(default_factory=list)
|
||||
previous_ai_analysis: list[Artifact] = field(default_factory=list)
|
||||
security_findings: list[Artifact] = field(default_factory=list)
|
||||
|
||||
# Raw graph for UI / debugging
|
||||
relationships_traversed: list[ArtifactRelationship] = field(default_factory=list)
|
||||
all_artifacts: list[Artifact] = field(default_factory=list)
|
||||
|
||||
semantic_hits: list[SemanticHit] = field(default_factory=list)
|
||||
|
||||
warnings: list[str] = field(default_factory=list)
|
||||
|
||||
def summary(self) -> dict:
|
||||
return {
|
||||
"seed_artifact_id": str(self.seed_artifact_id) if self.seed_artifact_id else None,
|
||||
"total_artifacts": len(self.all_artifacts),
|
||||
"total_relationships": len(self.relationships_traversed),
|
||||
"semantic_hits": len(self.semantic_hits),
|
||||
"by_type": {
|
||||
"requirements": len(self.requirements),
|
||||
"architecture_decisions": len(self.architecture_decisions),
|
||||
"tasks": len(self.tasks),
|
||||
"source_files": len(self.source_files),
|
||||
"documents": len(self.documents),
|
||||
"tests": len(self.tests),
|
||||
"builds": len(self.builds),
|
||||
"deployments": len(self.deployments),
|
||||
"releases": len(self.releases),
|
||||
"incidents": len(self.incidents),
|
||||
"previous_ai_analysis": len(self.previous_ai_analysis),
|
||||
"security_findings": len(self.security_findings),
|
||||
},
|
||||
"warnings": self.warnings,
|
||||
}
|
||||
|
||||
|
||||
class ContextEngine:
|
||||
"""Assembles contextual "fascicolo" given a seed (artifact, project, ...).
|
||||
|
||||
For the bootstrap phase we only accept explicit service-layer objects
|
||||
(relationship service + artifact service) so we can run the engine
|
||||
inside asf-core or standalone. No Mongo driver coupling here.
|
||||
"""
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
*,
|
||||
relationship_service, # duck-typed: for_artifact, outgoing, incoming
|
||||
artifact_service, # duck-typed: get, get_many, list
|
||||
semantic: SemanticStore | None = None,
|
||||
) -> None:
|
||||
self._rels = relationship_service
|
||||
self._arts = artifact_service
|
||||
self._semantic = semantic or InMemorySemanticStoreMock()
|
||||
|
||||
async def assemble(
|
||||
self,
|
||||
*,
|
||||
tenant_id: UUID,
|
||||
seed_artifact_id: Optional[UUID] = None,
|
||||
project_id: Optional[UUID] = None,
|
||||
user_query: Optional[str] = None,
|
||||
options: Optional[AssemblyOptions] = None,
|
||||
) -> ContextAssembly:
|
||||
opts = options or AssemblyOptions()
|
||||
result = ContextAssembly(
|
||||
seed_artifact_id=seed_artifact_id,
|
||||
project_id=project_id,
|
||||
tenant_id=tenant_id,
|
||||
)
|
||||
|
||||
visited_ids: set[UUID] = set()
|
||||
frontier: set[UUID] = set()
|
||||
|
||||
if seed_artifact_id is not None:
|
||||
frontier.add(seed_artifact_id)
|
||||
elif project_id is not None:
|
||||
arts = await self._arts.list(
|
||||
project_id=project_id, limit=opts.max_nodes
|
||||
)
|
||||
frontier = {a.id for a in arts[: opts.max_nodes]}
|
||||
result.all_artifacts.extend(arts)
|
||||
visited_ids.update(a.id for a in arts)
|
||||
else:
|
||||
result.warnings.append(
|
||||
"No seed_artifact_id nor project_id provided; context will be empty."
|
||||
)
|
||||
return result
|
||||
|
||||
# BFS walk
|
||||
for _ in range(opts.max_depth):
|
||||
current = list(frontier - visited_ids)
|
||||
if not current:
|
||||
break
|
||||
visited_ids.update(current)
|
||||
|
||||
# Load artifacts we haven't loaded yet
|
||||
need = [i for i in current if not any(a.id == i for a in result.all_artifacts)]
|
||||
if need:
|
||||
fetched = await self._arts.get_many(need)
|
||||
result.all_artifacts.extend(fetched)
|
||||
|
||||
for aid in current:
|
||||
if len(result.relationships_traversed) >= opts.max_nodes:
|
||||
break
|
||||
out = await self._rels.outgoing(
|
||||
aid, limit=max(50, opts.max_nodes // 4)
|
||||
)
|
||||
inc = await self._rels.incoming(
|
||||
aid, limit=max(50, opts.max_nodes // 4)
|
||||
)
|
||||
for rel in out + inc:
|
||||
result.relationships_traversed.append(rel)
|
||||
frontier.add(rel.source_artifact_id)
|
||||
frontier.add(rel.target_artifact_id)
|
||||
|
||||
# Ensure all referenced artifacts are loaded
|
||||
referenced: set[UUID] = set()
|
||||
for rel in result.relationships_traversed:
|
||||
referenced.add(rel.source_artifact_id)
|
||||
referenced.add(rel.target_artifact_id)
|
||||
already_have = {a.id for a in result.all_artifacts}
|
||||
missing = list(referenced - already_have)
|
||||
if missing:
|
||||
fetched = await self._arts.get_many(missing)
|
||||
result.all_artifacts.extend(fetched)
|
||||
|
||||
# Bucket artifacts by type
|
||||
_bucket(result)
|
||||
|
||||
# Optional semantic layer
|
||||
if opts.include_semantic and user_query:
|
||||
try:
|
||||
result.semantic_hits = list(
|
||||
await self._semantic.search(
|
||||
tenant_id=tenant_id,
|
||||
query=user_query,
|
||||
project_id=project_id,
|
||||
top_k=opts.semantic_top_k,
|
||||
)
|
||||
)
|
||||
except Exception: # noqa: BLE001
|
||||
await log.aexception("semantic_search_failed")
|
||||
result.warnings.append("Semantic search failed; using structural context only.")
|
||||
|
||||
return result
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# Internal helpers
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
|
||||
_TYPE_BUCKETS = {
|
||||
"requirement": "requirements",
|
||||
"architecture_decision": "architecture_decisions",
|
||||
"task": "tasks",
|
||||
"story": "tasks",
|
||||
"epic": "tasks",
|
||||
"source_file": "source_files",
|
||||
"document": "documents",
|
||||
"test": "tests",
|
||||
"build": "builds",
|
||||
"deployment": "deployments",
|
||||
"release": "releases",
|
||||
"incident": "incidents",
|
||||
"runtime_event": "incidents",
|
||||
"ai_analysis": "previous_ai_analysis",
|
||||
"security_finding": "security_findings",
|
||||
}
|
||||
|
||||
|
||||
def _bucket(r: ContextAssembly) -> None:
|
||||
for art in r.all_artifacts:
|
||||
t = art.type if isinstance(art.type, str) else art.type.value
|
||||
bucket = _TYPE_BUCKETS.get(t)
|
||||
if bucket is not None:
|
||||
getattr(r, bucket).append(art)
|
||||
@@ -0,0 +1,71 @@
|
||||
"""Semantic / vector retrieval abstraction (stub + in-memory mock).
|
||||
|
||||
A real implementation will talk to a vector DB / embedding store and return
|
||||
"semantically similar" artifacts. For bootstrap we keep the Protocol open and
|
||||
provide a deterministic mock.
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
from dataclasses import dataclass, field
|
||||
from typing import Protocol, Optional, Sequence
|
||||
from uuid import UUID
|
||||
|
||||
|
||||
@dataclass
|
||||
class SemanticHit:
|
||||
artifact_id: UUID
|
||||
score: float
|
||||
snippet: str = ""
|
||||
metadata: dict = field(default_factory=dict)
|
||||
|
||||
|
||||
class SemanticStore(Protocol):
|
||||
"""Future vector / embedding store interface."""
|
||||
|
||||
async def search(
|
||||
self,
|
||||
tenant_id: UUID,
|
||||
query: str,
|
||||
*,
|
||||
project_id: Optional[UUID] = None,
|
||||
top_k: int = 10,
|
||||
filters: Optional[dict] = None,
|
||||
) -> Sequence[SemanticHit]: ...
|
||||
|
||||
async def index_artifact(
|
||||
self,
|
||||
tenant_id: UUID,
|
||||
artifact_id: UUID,
|
||||
text: str,
|
||||
metadata: Optional[dict] = None,
|
||||
) -> None: ...
|
||||
|
||||
|
||||
class InMemorySemanticStoreMock(SemanticStore):
|
||||
"""Deterministic, no-op mock.
|
||||
|
||||
Useful for unit tests and for getting the Context Engine scaffolding up
|
||||
before a real vector DB is deployed. Always returns an empty hit list so
|
||||
callers gracefully degrade to structural context only.
|
||||
"""
|
||||
|
||||
async def search(
|
||||
self,
|
||||
tenant_id: UUID,
|
||||
query: str,
|
||||
*,
|
||||
project_id: Optional[UUID] = None,
|
||||
top_k: int = 10,
|
||||
filters: Optional[dict] = None,
|
||||
) -> Sequence[SemanticHit]:
|
||||
return tuple()
|
||||
|
||||
async def index_artifact(
|
||||
self,
|
||||
tenant_id: UUID,
|
||||
artifact_id: UUID,
|
||||
text: str,
|
||||
metadata: Optional[dict] = None,
|
||||
) -> None: # pragma: no cover - mock side effect
|
||||
return None
|
||||
Reference in New Issue
Block a user