Use haiku.rag as telemtry scope

haiku_rag_slim/haiku/rag/chat/app.py

@@ -32,6 +32,7 @@ from haiku.rag.client import HaikuRAG
 from haiku.rag.config import get_config
 from haiku.rag.skills.analysis import AnalysisState
 from haiku.rag.skills.rag import RAGState, get_agent_preamble
+from haiku.rag.telemetry import configure as configure_telemetry
 from haiku.skills.agent import (
     SkillToolset,
     run_agui_stream,
@@ -39,13 +40,7 @@ from haiku.skills.agent import (
 from haiku.skills.models import Skill
 from haiku.skills.prompts import build_system_prompt
 
-try:
-    import logfire
-
-    logfire.configure(send_to_logfire="if-token-present", console=False)
-    logfire.instrument_pydantic_ai()
-except ImportError:
-    pass
+configure_telemetry()
 
 if TYPE_CHECKING:
     from textual.app import ComposeResult

haiku_rag_slim/haiku/rag/cli.py

@@ -134,18 +134,10 @@ def main(
     # Configure logging for CLI context
     configure_cli_logging()
 
-    # Configure logfire (only sends data if token is present)
-    try:
-        import logfire
+    from haiku.rag.telemetry import configure as configure_telemetry
 
-        is_production = get_config().environment != "development"
-        logfire.configure(
-            send_to_logfire="if-token-present",
-            console=False if is_production else None,
-        )
-        logfire.instrument_pydantic_ai()
-    except Exception:  # pragma: no cover
-        pass
+    is_production = get_config().environment != "development"
+    configure_telemetry(console=False if is_production else None)
 
     if get_config().environment != "development":
         # Suppress warnings in production

haiku_rag_slim/haiku/rag/client/documents.py

@@ -3,8 +3,6 @@ from pathlib import Path
 from typing import TYPE_CHECKING
 from urllib.parse import unquote, urlparse
 
-import logfire
-
 from haiku.rag.client.exceptions import UnsupportedSourceError
 from haiku.rag.client.processing import (
     ensure_chunks_embedded,
@@ -16,6 +14,7 @@ from haiku.rag.ingester.sources import FetchResult, resolve_fetcher
 from haiku.rag.store.models.chunk import Chunk
 from haiku.rag.store.models.document import Document
 from haiku.rag.store.models.document_item import extract_items
+from haiku.rag.telemetry import logfire
 
 if TYPE_CHECKING:
     from docling_core.types.doc.document import DoclingDocument

haiku_rag_slim/haiku/rag/converters/pdf_split.py

@@ -6,10 +6,10 @@ from collections.abc import Generator
 from pathlib import Path
 from typing import TYPE_CHECKING
 
-import logfire
 import pypdfium2 as pdfium
 
 from haiku.rag.client.exceptions import UnsupportedSourceError
+from haiku.rag.telemetry import logfire
 
 # pypdfium2 wraps libpdfium, which has global C state and is not thread-safe.
 # Multiple workers calling iter_pdf_slices concurrently race on that state

haiku_rag_slim/haiku/rag/ingester/cli.py

@@ -50,28 +50,12 @@ def main(
     _load_config_with_override(config)
 
 
-def _configure_logfire() -> None:
-    """Logfire emits spans only when LOGFIRE_TOKEN is set; otherwise it
-    stays silent. Console output is disabled in either case so span lines
-    don't interleave with the ingester's own RichHandler logs — telemetry
-    lives in the Logfire UI."""
-    try:
-        import logfire
-
-        logfire.configure(
-            service_name="haiku-ingester",
-            send_to_logfire="if-token-present",
-            console=False,
-        )
-        logfire.instrument_pydantic_ai()
-    except Exception:  # pragma: no cover
-        pass
-
-
 def cli() -> None:
     """Entry point that translates store-state errors into a clean exit."""
+    from haiku.rag.telemetry import configure as configure_telemetry
+
     configure_cli_logging()
-    _configure_logfire()
+    configure_telemetry(service_name="haiku-ingester")
     try:
         _cli()
     except (MigrationRequiredError, ReadOnlyError) as e:

haiku_rag_slim/haiku/rag/ingester/pollers/base.py

@@ -2,8 +2,6 @@ import asyncio
 import logging
 from datetime import UTC, datetime
 
-import logfire
-
 from haiku.rag.config import SourceConfig
 from haiku.rag.ingester.pollers.circuit_breaker import CircuitBreaker
 from haiku.rag.ingester.queue.models import JobOp
@@ -13,6 +11,7 @@ from haiku.rag.ingester.sources.base import (
     SourceEvent,
     SourceEventKind,
 )
+from haiku.rag.telemetry import get_context, logfire
 
 logger = logging.getLogger(__name__)
 
@@ -29,7 +28,7 @@ def _enqueue_extra(cfg: SourceConfig) -> dict | None:
     headers = getattr(cfg, "headers", None)
     if headers:
         extra["headers"] = dict(headers)
-    carrier = logfire.get_context()
+    carrier = get_context()
     if carrier:
         extra["_otel"] = dict(carrier)
     return extra or None

haiku_rag_slim/haiku/rag/ingester/pollers/fs.py

@@ -3,12 +3,12 @@ import logging
 from pathlib import Path
 from typing import TYPE_CHECKING
 
-import logfire
 from watchfiles import Change, awatch
 
 from haiku.rag.ingester.pollers.base import BasePoller, _enqueue_extra
 from haiku.rag.ingester.queue.models import JobOp
 from haiku.rag.ingester.sources.filter import FileFilter
+from haiku.rag.telemetry import logfire
 
 if TYPE_CHECKING:
     from haiku.rag.config import FSSourceConfig

haiku_rag_slim/haiku/rag/ingester/workers/pipeline.py

@@ -3,12 +3,12 @@ from contextlib import nullcontext
 from typing import TYPE_CHECKING
 
 import httpx
-import logfire
 from pydantic import BaseModel
 
 from haiku.rag.client.exceptions import UnsupportedSourceError
 from haiku.rag.ingester.exceptions import PermanentError, TransientError
 from haiku.rag.ingester.queue.models import Job, JobOp
+from haiku.rag.telemetry import attach_context, logfire
 
 if TYPE_CHECKING:
     from haiku.rag.client import HaikuRAG
@@ -75,7 +75,7 @@ async def run_job(client: "HaikuRAG", job: Job) -> JobResult:
     # Restore the poller's trace context (if any) so the job span nests
     # under the `ingester.poller.sweep` that enqueued it.
     parent_ctx = extra.get("_otel")
-    attach = logfire.attach_context(parent_ctx) if parent_ctx else nullcontext()
+    attach = attach_context(parent_ctx) if parent_ctx else nullcontext()
 
     with (
         attach,

haiku_rag_slim/haiku/rag/telemetry.py

@@ -0,0 +1,47 @@
+from typing import Literal
+
+from logfire import Logfire, attach_context, get_context
+
+# Scoped Logfire instance — every span emitted through `logfire.span(...)`
+# on this object carries `instrumentation_scope.name = "haiku.rag"` instead
+# of the default "logfire". The scope is OTel's identifier for *which
+# library* produced the span, separate from `service.name` which is the
+# running process. Downstream consumers (Logfire UI saved views, OTel
+# collectors, alert rules) can then filter on `scope.name = haiku.rag`
+# rather than catching every span the SDK ever exports.
+#
+# Cross-library instrumentations (pydantic-ai, FastAPI, OpenAI) keep their
+# own scopes — this only retags the spans WE write.
+logfire = Logfire(otel_scope="haiku.rag")
+
+
+def configure(
+    *,
+    service_name: str | None = None,
+    console: Literal[False] | None = False,
+) -> None:
+    """Configure Logfire and enable pydantic-ai instrumentation for the
+    running process. Each CLI entry point calls this once at startup.
+    Silently no-ops on failure so a missing/misconfigured LOGFIRE_TOKEN
+    never crashes the app.
+
+    - service_name: identifies the process in the Logfire UI (e.g.
+      "haiku-ingester"). Falls back to logfire's default when None.
+    - console: False (default) suppresses span lines on stderr so they
+      don't interleave with RichHandler logs. Pass None to let logfire
+      decide (its own default applies).
+    """
+    try:
+        import logfire as _lf
+
+        _lf.configure(
+            service_name=service_name,
+            send_to_logfire="if-token-present",
+            console=console,
+        )
+        _lf.instrument_pydantic_ai()
+    except Exception:  # pragma: no cover
+        pass
+
+
+__all__ = ["attach_context", "configure", "get_context", "logfire"]