unique-sdk 2026.30.0.dev7__tar.gz → 2026.30.0.dev9__tar.gz

{unique_sdk-2026.30.0.dev7 → unique_sdk-2026.30.0.dev9}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: unique-sdk
-Version: 2026.30.0.dev7
+Version: 2026.30.0.dev9
 Summary: 
 Author: Martin Fadler, Konstantin Krauss, Andreas Hauri
 Author-email: Martin Fadler <martin.fadler@unique.ch>, Konstantin Krauss <konstantin@unique.ch>, Andreas Hauri <andreas@unique.ch>

{unique_sdk-2026.30.0.dev7 → unique_sdk-2026.30.0.dev9}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "unique_sdk"
-version = "2026.30.0.dev7"
+version = "2026.30.0.dev9"
 description = ""
 readme = "README.md"
 license = { text = "MIT" }

{unique_sdk-2026.30.0.dev7 → unique_sdk-2026.30.0.dev9}/unique_sdk/cli/identity.py

@@ -25,7 +25,6 @@ from pathlib import Path
 
 TURN_IDENTITY_ENV_VAR = "UNIQUE_TURN_IDENTITY_FILE"
 MESSAGE_ID_ENV_VAR = "UNIQUE_MESSAGE_ID"
-CHAT_ID_ENV_VAR = "UNIQUE_CHAT_ID"
 
 
 class TurnIdentityError(ValueError):
@@ -36,23 +35,13 @@ class TurnIdentityError(ValueError):
 class TurnIdentity:
     """Parsed contents of the per-turn identity file.
 
-    Mirrors the JSON contract written by the platform runner. Only
-    ``message_id`` is mandatory; the remaining fields are informational.
+    The runner writes a richer JSON object (chat/user/company/assistant IDs,
+    turn counter), but the CLI only consumes ``message_id`` — the one value
+    that goes stale in a persistent process environment. Extra keys are
+    ignored; parse a field here only once the CLI actually relies on it.
     """
 
     message_id: str
-    chat_id: str | None = None
-    user_id: str | None = None
-    company_id: str | None = None
-    assistant_id: str | None = None
-    turn: int | None = None
-
-
-def _optional_str(payload: dict[str, object], key: str) -> str | None:
-    value = payload.get(key)
-    if isinstance(value, str) and value.strip():
-        return value.strip()
-    return None
 
 
 def read_turn_identity(
@@ -99,15 +88,7 @@ def read_turn_identity(
             "'message_id' string"
         )
 
-    raw_turn = payload.get("turn")
-    return TurnIdentity(
-        message_id=message_id.strip(),
-        chat_id=_optional_str(payload, "chat_id"),
-        user_id=_optional_str(payload, "user_id"),
-        company_id=_optional_str(payload, "company_id"),
-        assistant_id=_optional_str(payload, "assistant_id"),
-        turn=raw_turn if isinstance(raw_turn, int) else None,
-    )
+    return TurnIdentity(message_id=message_id.strip())
 
 
 def resolve_message_id(explicit: str | None = None) -> str | None:
@@ -129,22 +110,3 @@ def resolve_message_id(explicit: str | None = None) -> str | None:
     if env_id is not None and env_id.strip():
         return env_id.strip()
     return None
-
-
-def resolve_chat_id(explicit: str | None = None) -> str | None:
-    """Resolve chat ID with the same file-then-env precedence as message ID.
-
-    Explicit flag wins. The turn-identity file is consulted next (and fails
-    loudly when configured but unusable). Falls back to ``$UNIQUE_CHAT_ID``.
-    """
-    if explicit is not None and str(explicit).strip():
-        return str(explicit).strip()
-
-    identity = read_turn_identity()
-    if identity is not None and identity.chat_id is not None:
-        return identity.chat_id
-
-    env_id = os.environ.get(CHAT_ID_ENV_VAR)
-    if env_id is not None and env_id.strip():
-        return env_id.strip()
-    return None

{unique_sdk-2026.30.0.dev7 → unique_sdk-2026.30.0.dev9}/unique_sdk/utils/chat_in_space.py

@@ -1,3 +1,5 @@
+from __future__ import annotations
+
 import asyncio
 import warnings
 from collections.abc import Awaitable, Callable, Sequence
@@ -11,6 +13,17 @@ from unique_sdk.utils.file_io import (
 )
 
 
+def get_message_invocations(message: Space.Message) -> list[dict[str, Any]]:
+    """Per-LLM-invocation stats for a completed Space message, read from
+    `debugInfo.llm_invocations` (a plain list, one entry per LLM call).
+
+    Each entry has `modelName`, `tokenUsage`, and `source` keys.
+    """
+    debug_info = message.get("debugInfo") or {}
+    llm_invocations = debug_info.get("llm_invocations")
+    return llm_invocations if isinstance(llm_invocations, list) else []
+
+
 async def send_message_and_wait_for_completion(
     user_id: str,
     company_id: str,
@@ -23,10 +36,10 @@ async def send_message_and_wait_for_completion(
     poll_interval: float = 1.0,
     max_wait: float = 60.0,
     stop_condition: Literal["stoppedStreamingAt", "completedAt"] = "stoppedStreamingAt",
-    correlation: "Space.Correlation | None" = None,
+    correlation: Space.Correlation | None = None,
     auto_approve_elicitation: bool | None = None,
-    on_message_update: Callable[["Space.Message"], Awaitable[None]] | None = None,
-) -> "Space.Message":
+    on_message_update: Callable[[Space.Message], Awaitable[None]] | None = None,
+) -> Space.Message:
     """
     Sends a prompt asynchronously and polls for completion. (until stoppedStreamingAt is not None)
 
@@ -42,6 +55,13 @@ async def send_message_and_wait_for_completion(
         poll_interval: Seconds between polls.
         max_wait: Maximum seconds to wait for completion.
         stop_condition: Defines when to expect a response back, when the assistant stop streaming or when it completes the message. (default: "stoppedStreamingAt")
+            Note: `debugInfo.llm_invocations` (see `get_message_invocations`) is written
+            near the very end of the orchestrator's run -- after postprocessors and
+            evaluations, which happen after the visible text stream ends. With the default
+            `"stoppedStreamingAt"`, or even `"completedAt"` (set slightly before the debugInfo
+            write completes), the returned message's invocations can still be incomplete
+            immediately after this function returns. Re-fetch the message after a short delay
+            if you need the full invocation list.
         correlation: Optional correlation data to link this message to a parent message in another chat.
             Should contain: parentMessageId, parentChatId, parentAssistantId.
         auto_approve_elicitation: When True, automatically approves elicitation requests during
@@ -91,6 +111,11 @@ async def send_message_and_wait_for_completion(
                 last_update_signature = update_signature
         if answer.get(stop_condition) is not None:
             try:
+                # debugInfo/llm_invocations is written via update_debug_info_async(),
+                # which always targets the original USER message, not the
+                # assistant reply (assistant=False in
+                # ChatService._construct_message_modify_params) — re-fetch by
+                # `message_id`, not the polled assistant message's own id.
                 user_message = await Message.retrieve_async(
                     user_id, company_id, message_id, chatId=chat_id
                 )
@@ -116,7 +141,7 @@ async def chat_against_file(
     poll_interval: float = 1.0,
     max_wait: float = 60.0,
     should_delete_chat: bool = True,
-) -> "Space.Message":
+) -> Space.Message:
     """
     Chat against a file by uploading it, sending a message and waiting for a reply.
     Args: