modelstat-sdk 0.0.1__tar.gz → 0.0.2__tar.gz

{modelstat_sdk-0.0.1 → modelstat_sdk-0.0.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: modelstat-sdk
-Version: 0.0.1
+Version: 0.0.2
 Summary: Privacy-first SDK for modelstat — wrap your backend LLM calls and ship redacted usage to a local daemon or the modelstat server, without touching live-request latency.
 Project-URL: Homepage, https://modelstat.ai
 Project-URL: Repository, https://github.com/modelstat/modelstat
@@ -137,13 +137,22 @@ cfg = Config("msk_live_…", "raw_sdk_openai").with_remote(
 )
 ```
 
+## Taxonomy auto-detection (off by default)
+
+modelstat can auto-detect a work-type *taxonomy* over your sessions, but that's tuned for interactive coding sessions — backend LLM usage usually isn't. So for the SDK taxonomy is **off by default**: every batch ships an explicit `auto_taxonomy: false`. Opt in with the config flag:
+
+```python
+cfg = Config("msk_live_…", "raw_sdk_openai")
+cfg.auto_taxonomy = True  # force server-side taxonomy auto-detection on
+```
+
 ## Privacy floor (always on)
 
 Before any bytes leave the SDK process — in **every** mode — an in-process redaction floor scrubs secrets (provider keys, tokens, JWTs, PEM blocks, DB passwords, …), emails, and absolute home paths. "Raw" mode means *full turns*, not *leaked credentials* — the floor still runs. Tool calls ship only hashes, byte sizes, and allowlisted command verbs — never raw args, results, paths, or command text.
 
 What the floor redacts: Anthropic / OpenAI / Google / AWS / GitHub / Slack / Stripe / Discord keys and tokens, JWTs, PEM private-key blocks, modelstat device secrets, generic `NAME_KEY=value` env secrets (the name is kept, the value is dropped), `Bearer` tokens, database-URL passwords, lone 40-char AWS-style secret blobs, email addresses, and absolute `/Users/…`, `/home/…`, and `C:\Users\…` paths.
 
-## What's live today (v0.0.1)
+## What's live today (v0.0.2)
 
 Early release — the honest state, so nothing surprises you:
 

{modelstat_sdk-0.0.1 → modelstat_sdk-0.0.2}/README.md

@@ -111,13 +111,22 @@ cfg = Config("msk_live_…", "raw_sdk_openai").with_remote(
 )
 ```
 
+## Taxonomy auto-detection (off by default)
+
+modelstat can auto-detect a work-type *taxonomy* over your sessions, but that's tuned for interactive coding sessions — backend LLM usage usually isn't. So for the SDK taxonomy is **off by default**: every batch ships an explicit `auto_taxonomy: false`. Opt in with the config flag:
+
+```python
+cfg = Config("msk_live_…", "raw_sdk_openai")
+cfg.auto_taxonomy = True  # force server-side taxonomy auto-detection on
+```
+
 ## Privacy floor (always on)
 
 Before any bytes leave the SDK process — in **every** mode — an in-process redaction floor scrubs secrets (provider keys, tokens, JWTs, PEM blocks, DB passwords, …), emails, and absolute home paths. "Raw" mode means *full turns*, not *leaked credentials* — the floor still runs. Tool calls ship only hashes, byte sizes, and allowlisted command verbs — never raw args, results, paths, or command text.
 
 What the floor redacts: Anthropic / OpenAI / Google / AWS / GitHub / Slack / Stripe / Discord keys and tokens, JWTs, PEM private-key blocks, modelstat device secrets, generic `NAME_KEY=value` env secrets (the name is kept, the value is dropped), `Bearer` tokens, database-URL passwords, lone 40-char AWS-style secret blobs, email addresses, and absolute `/Users/…`, `/home/…`, and `C:\Users\…` paths.
 
-## What's live today (v0.0.1)
+## What's live today (v0.0.2)
 
 Early release — the honest state, so nothing surprises you:
 

{modelstat_sdk-0.0.1 → modelstat_sdk-0.0.2}/src/modelstat/__init__.py

@@ -46,7 +46,7 @@ from .config import DEFAULT_DAEMON_URL, Config, Mode, RedactionPolicy
 from .redact import Redacted, redact
 from .transport import FakeTransport, HttpTransport, Transport, TransportError
 from .wire import (
-    BillingMode,
+    PricingMode,
     EventKind,
     GitContext,
     IngestBatch,
@@ -86,7 +86,7 @@ __all__ = [
     "TokenUsage",
     "GitContext",
     "EventKind",
-    "BillingMode",
+    "PricingMode",
     "ToolCallStatus",
     "content_hash",
     "source_event_id",

{modelstat_sdk-0.0.1 → modelstat_sdk-0.0.2}/src/modelstat/_version.py

@@ -5,4 +5,4 @@ Read both at runtime (to build ``Config.client_version`` -> the wire
 ``[tool.hatch.version]``), so the two can never drift.
 """
 
-__version__ = "0.0.1"
+__version__ = "0.0.2"

{modelstat_sdk-0.0.1 → modelstat_sdk-0.0.2}/src/modelstat/capture.py

@@ -19,7 +19,7 @@ from . import wire
 from .config import Config, RedactionPolicy
 from .redact import redact
 from .wire import (
-    BillingMode,
+    PricingMode,
     EventKind,
     GitContext,
     IngestBatch,
@@ -83,7 +83,7 @@ class LlmCall:
     completion: Optional[str] = None
     cwd: Optional[str] = None
     git: Optional[GitContext] = None
-    billing: Optional[BillingMode] = None
+    pricing_mode: Optional[PricingMode] = None
     tool_calls: List[ToolCallInput] = field(default_factory=list)
 
     # ---- chainable builder helpers (ergonomic, mirror the Rust builder) -----
@@ -193,7 +193,7 @@ def _event_from_call(
         cwd=call.cwd,
         git=call.git,
         duration_ms=call.duration_ms,
-        billing=call.billing,
+        pricing_mode=call.pricing_mode,
         content_excerpt=_build_excerpt(cfg, call),
     )
 
@@ -260,5 +260,8 @@ def build_batch(
         daemon_version=cfg.client_version,
         events=events,
         tool_calls=tool_calls,
+        # Always send an explicit value reflecting the config so backend usage is
+        # off-by-default but users can opt in.
+        auto_taxonomy=cfg.auto_taxonomy,
     )
     return batch, seq

{modelstat_sdk-0.0.1 → modelstat_sdk-0.0.2}/src/modelstat/config.py

@@ -108,6 +108,12 @@ class Config:
     flush_interval: float = 2.0
     # Flush eagerly once this many records are buffered.
     flush_max_batch: int = 256
+    # Whether the server should run taxonomy auto-detection on batches from this
+    # client. Ships as the wire ``auto_taxonomy`` field. Defaults to ``False``
+    # for SDK/backend integrations -- backend LLM usage isn't interactive
+    # work-sessions, so taxonomy is **off by default**; set it to ``True`` to opt
+    # in.
+    auto_taxonomy: bool = False
 
     def __post_init__(self) -> None:
         # The wire field is constrained to 1..=40 chars; keep the SDK honest so

{modelstat_sdk-0.0.1 → modelstat_sdk-0.0.2}/src/modelstat/wire.py

@@ -36,7 +36,7 @@ import blake3
 __all__ = [
     "TokenUsage",
     "EventKind",
-    "BillingMode",
+    "PricingMode",
     "ToolCallStatus",
     "GitContext",
     "RawEvent",
@@ -118,7 +118,7 @@ class EventKind(str, Enum):
     SUMMARY = "summary"
 
 
-class BillingMode(str, Enum):
+class PricingMode(str, Enum):
     """How the provider billed the call."""
 
     SUBSCRIPTION = "subscription"
@@ -181,7 +181,7 @@ class RawEvent:
     cwd: Optional[str] = None
     git: Optional[GitContext] = None
     duration_ms: Optional[int] = None
-    billing: Optional[BillingMode] = None
+    pricing_mode: Optional[PricingMode] = None
     # Redacted excerpt used to build summaries downstream. Capped at 320 chars
     # in the standard (floor-redacted) path; carries the full redacted turns in
     # remote-raw mode, where the server summarizes.
@@ -206,8 +206,8 @@ class RawEvent:
             out["git"] = self.git.to_dict()
         if self.duration_ms is not None:
             out["duration_ms"] = self.duration_ms
-        if self.billing is not None:
-            out["billing"] = self.billing.value
+        if self.pricing_mode is not None:
+            out["pricing_mode"] = self.pricing_mode.value
         if self.content_excerpt is not None:
             out["content_excerpt"] = self.content_excerpt
         return out
@@ -292,6 +292,12 @@ class IngestBatch:
     daemon_version: str
     events: List[RawEvent] = field(default_factory=list)
     tool_calls: List[ToolCallWire] = field(default_factory=list)
+    # Per-batch taxonomy auto-detection toggle. ``None`` = server default
+    # (taxonomy auto/on); ``False`` = skip taxonomy auto-detection for this
+    # batch; ``True`` = force it on. SDK/backend integrations default this to
+    # ``False`` (backend LLM usage isn't interactive work-sessions). Included in
+    # ``to_dict()`` only when not None.
+    auto_taxonomy: Optional[bool] = None
 
     def to_dict(self) -> Dict[str, Any]:
         out: Dict[str, Any] = {
@@ -303,6 +309,9 @@ class IngestBatch:
         # Omit ``tool_calls`` entirely when empty (do NOT send an empty list).
         if self.tool_calls:
             out["tool_calls"] = [t.to_dict() for t in self.tool_calls]
+        # Optional key -- omit when None (never emit null).
+        if self.auto_taxonomy is not None:
+            out["auto_taxonomy"] = self.auto_taxonomy
         return out
 
 

{modelstat_sdk-0.0.1 → modelstat_sdk-0.0.2}/tests/test_capture.py

@@ -115,6 +115,19 @@ class TestBuildBatch(unittest.TestCase):
         batch, _ = build_batch(cfg(), [call], 0)
         self.assertNotIn("tool_calls", batch.to_dict())
 
+    def test_auto_taxonomy_defaults_off_and_opts_in(self) -> None:
+        # Default config: taxonomy off -> explicit ``auto_taxonomy: false``.
+        batch, _ = build_batch(cfg(), [LlmCall("openai", "sess_1")], 0)
+        self.assertEqual(batch.auto_taxonomy, False)
+        self.assertEqual(batch.to_dict()["auto_taxonomy"], False)
+
+        # Opt in: flag True -> wire ``auto_taxonomy: true``.
+        on = cfg()
+        on.auto_taxonomy = True
+        batch, _ = build_batch(on, [LlmCall("openai", "sess_1")], 0)
+        self.assertEqual(batch.auto_taxonomy, True)
+        self.assertEqual(batch.to_dict()["auto_taxonomy"], True)
+
     def test_raw_mode_sends_full_untruncated_turns_still_floor_redacted(
         self,
     ) -> None:

{modelstat_sdk-0.0.1 → modelstat_sdk-0.0.2}/tests/test_wire.py

@@ -7,7 +7,7 @@ import unittest
 from datetime import datetime, timezone
 
 from modelstat.wire import (
-    BillingMode,
+    PricingMode,
     EventKind,
     RawEvent,
     TokenUsage,
@@ -57,13 +57,13 @@ class TestSerialization(unittest.TestCase):
             tokens=TokenUsage(input=10, output=5),
             model="gpt-x",
             duration_ms=1200,
-            billing=BillingMode.API,
+            pricing_mode=PricingMode.API,
             content_excerpt="hello",
         )
         j = ev.to_dict()
         self.assertEqual(j["kind"], "assistant_message")
         self.assertEqual(j["agent"], "raw_sdk_openai")
-        self.assertEqual(j["billing"], "api")
+        self.assertEqual(j["pricing_mode"], "api")
         self.assertEqual(j["tokens"]["input"], 10)
         # Tokens object always carries all five classes.
         self.assertEqual(