Build Agentic workflows, RAG pipelines, with rich AI model and tool support on one collaborative workspace. Deploy on cloud, VPC, or self-hosted, so teams move from prototype to production without rebuilding the stack.
Python agentagentic-aiagentic-frameworkagentic-workflowaiautomationclaudedeepseekgenaigptllmlow-codemcpnextjsno-codeopenaipythonskillsworkflow
"""Small, typed Celery locators. Runtime scope is reloaded from the KnowledgeFS database."""
from typing import Literal
from uuid import UUID
from pydantic import BaseModel, ConfigDict, Field, model_validator
DOCUMENT_QUEUE = "knowledge_fs_document"
PRIORITY_DOCUMENT_QUEUE = "knowledge_fs_priority_document"
SOURCE_QUEUE = "knowledge_fs_source"
RESEARCH_QUEUE = "knowledge_fs_research"
MAINTENANCE_QUEUE = "knowledge_fs_maintenance"
DISPATCH_QUEUE = "knowledge_fs_dispatch"
DELIVERY_TASK = "tasks.knowledge_fs_background_tasks.execute_delivery"
OPERATION_TASK = "tasks.knowledge_fs_background_tasks.execute_operation"
# Database-owned bounded sweeps are separate from document broker messages. In particular a source
# workflow must never occupy the only worker capable of completing its child compilation.
OPERATION_QUEUES = {
"document.dispatch": DISPATCH_QUEUE,
"document.reconcile": MAINTENANCE_QUEUE,
"document.semantic": DOCUMENT_QUEUE,
"page-index.repair": DOCUMENT_QUEUE,
"page-index.upgrade": MAINTENANCE_QUEUE,
"legacy.bootstrap": MAINTENANCE_QUEUE,
"profile.migrate": DOCUMENT_QUEUE,
"profile.backfill": MAINTENANCE_QUEUE,
"fts.backfill": MAINTENANCE_QUEUE,
"research.execute": RESEARCH_QUEUE,
"deletion.dispatch": DISPATCH_QUEUE,
"deletion.execute": MAINTENANCE_QUEUE,
"upload.cleanup": MAINTENANCE_QUEUE,
"vector.cleanup": MAINTENANCE_QUEUE,
"source.execute": SOURCE_QUEUE,
"source.schedule": DISPATCH_QUEUE,
"source.preview": SOURCE_QUEUE,
"source.legacy-sync": SOURCE_QUEUE,
"quality.replay": RESEARCH_QUEUE,
}
class CompilationLocator(BaseModel):
model_config = ConfigDict(extra="forbid")
attemptId: UUID
class FindabilityLocator(BaseModel):
model_config = ConfigDict(extra="forbid")
compilationAttemptId: UUID
publicationFingerprint: str = Field(min_length=1, max_length=128)
class BackgroundJobPayload(BaseModel):
model_config = ConfigDict(extra="forbid")
id: UUID
type: Literal["document.compile", "quality.page-index-findability"]
payload: CompilationLocator | FindabilityLocator
attempts: int = Field(default=1, ge=1, le=100_000, strict=True)
runAfter: int | None = Field(default=None, ge=0, le=2**53 - 1, strict=True)
# Publisher routing metadata must never become trusted worker execution scope.
priority: Literal["normal", "high"] = Field(default="normal", exclude=True)
@model_validator(mode="after")
def validate_locator_kind(self) -> "BackgroundJobPayload":
expected = CompilationLocator if self.type == "document.compile" else FindabilityLocator
if not isinstance(self.payload, expected):
raise ValueError("The locator does not match the background job type")
return self