lambda-erp 0.1.17__tar.gz → 0.1.19__tar.gz

{lambda_erp-0.1.17 → lambda_erp-0.1.19}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lambda-erp
-Version: 0.1.17
+Version: 0.1.19
 Summary: Core ERP logic - accounting, sales, purchasing, inventory
 Author: TORUS INVESTMENTS AG
 License-Expression: MIT

{lambda_erp-0.1.17 → lambda_erp-0.1.19}/api/chat.py

@@ -9,9 +9,12 @@ Supports multiple chat sessions, each with their own history and auto-generated
 """
 
 import asyncio
+import base64
+import io
 import json
 import logging
 import os
+import re
 import traceback
 import uuid
 from datetime import date
@@ -29,7 +32,7 @@ from api.demo_limits import (
     is_demo_role,
     limiter as demo_limiter,
 )
-from api.providers import cost_of_anthropic_call, cost_of_openai_call
+from api.providers import cost_of_anthropic_call, cost_of_openai_call, cost_of_transcription
 from api.routers.masters import create_master_record, update_master_record, MASTER_IDENTITY_ALIAS
 from lambda_erp.database import get_db
 from lambda_erp.utils import flt, now, nowdate
@@ -533,7 +536,22 @@ TOOLS = [
         "type": "function",
         "function": {
             "name": "cancel_document",
-            "description": "Cancel a submitted document (reverses GL/stock entries). Only works on submitted documents (docstatus=1). To void a draft, submit it first then cancel it. There is no delete — cancel is the only way to void a document.",
+            "description": "Cancel a SUBMITTED document (reverses its GL/stock entries). Only works on submitted documents (docstatus=1). For an unwanted DRAFT (docstatus=0), use discard_document instead — do NOT submit a draft just to cancel it.",
+            "parameters": {
+                "type": "object",
+                "properties": {
+                    "doctype": {"type": "string", "enum": DOCUMENT_SLUGS},
+                    "name": {"type": "string"},
+                },
+                "required": ["doctype", "name"],
+            },
+        },
+    },
+    {
+        "type": "function",
+        "function": {
+            "name": "discard_document",
+            "description": "Void/discard an unwanted DRAFT document (docstatus=0). This is a soft delete: the record is kept for the audit trail but flagged 'Discarded' and hidden from default lists. It has no financial or stock impact (a draft never posted anything). Only works on drafts — a submitted document must be cancelled instead. There is no hard delete in the system.",
             "parameters": {
                 "type": "object",
                 "properties": {
@@ -1051,6 +1069,10 @@ def _handle_cancel_document(args):
     return services.cancel_document(args["doctype"], args["name"])
 
 
+def _handle_discard_document(args):
+    return services.discard_document(args["doctype"], args["name"])
+
+
 def _handle_convert_document(args):
     return services.convert_document(args["doctype"], args["name"], args["target_doctype"])
 
@@ -1384,6 +1406,7 @@ TOOL_HANDLERS = {
     "update_document": _handle_update_document,
     "submit_document": _handle_submit_document,
     "cancel_document": _handle_cancel_document,
+    "discard_document": _handle_discard_document,
     "convert_document": _handle_convert_document,
     "search_masters": _handle_search_masters,
     "create_master": _handle_create_master,
@@ -2306,6 +2329,65 @@ async def run_thinking_loop(
     await on_event({"type": "complete", "content": content})
 
 
+# ---------------------------------------------------------------------------
+# Speech-to-text (voice → text)
+# ---------------------------------------------------------------------------
+
+# Domain-vocabulary biasing for the transcription `prompt`. A bare term list,
+# NO language constraint — the spoken language is auto-detected, and a
+# locale-derived hint would mis-transcribe users who speak another language
+# than their UI. Swiss German transcribes to Standard German on its own, which
+# is what the thinking loop wants. These are the ERP terms short utterances
+# otherwise mangle.
+VOICE_TRANSCRIBE_PROMPT = (
+    "Lambda, ERP, Offerte, Angebot, Auftrag, Rechnung, Gutschrift, Beleg, "
+    "Lieferschein, Bestellung, Zahlung, Buchung, Kunde, Lieferant, Artikel, "
+    "Lager, Konto, Kostenstelle, quotation, sales order, invoice, credit note, "
+    "purchase order, payment, journal entry, customer, supplier, item, warehouse"
+)
+
+# Below this the API rejects the clip as "too short"; we treat it as a no-op
+# (the user tapped the mic by accident) rather than an error.
+_MIN_TRANSCRIBE_BYTES = 1500  # ~1s of webm/opus at the ~1.5 KB/s we assume below
+
+
+def transcribe_audio(audio_bytes: bytes, audio_format: str, api_key: str) -> str:
+    """Transcribe audio to text via OpenAI gpt-4o-transcribe.
+
+    Returns the transcript, or "" when the model detects silence/near-silence
+    (empty result or a prompt-echo hallucination on a near-silent clip).
+    Raises on an actual API error so the caller can surface it.
+    """
+    client = OpenAI(api_key=api_key, timeout=httpx.Timeout(60.0, connect=10.0))
+
+    audio_file = io.BytesIO(audio_bytes)
+    audio_file.name = f"voice.{audio_format or 'webm'}"
+
+    transcription = client.audio.transcriptions.create(
+        model="gpt-4o-transcribe",
+        file=audio_file,
+        response_format="json",
+        prompt=VOICE_TRANSCRIBE_PROMPT,
+    )
+
+    text = (getattr(transcription, "text", "") or "").strip()
+    if len(text) < 2:
+        return ""
+
+    # Prompt-echo guard: on silent/noise audio, prompt-biased transcription
+    # tends to hallucinate the biasing vocabulary back. If (nearly) every word
+    # of the transcript is drawn from the prompt itself, it's an echo, not
+    # speech — real utterances carry verbs/objects outside the glossary.
+    prompt_tokens = set(re.findall(r"[\w'-]+", VOICE_TRANSCRIBE_PROMPT.lower()))
+    text_tokens = re.findall(r"[\w'-]+", text.lower())
+    if text_tokens:
+        coverage = sum(1 for tok in text_tokens if tok in prompt_tokens) / len(text_tokens)
+        if coverage >= 0.8:
+            return ""
+
+    return text
+
+
 # ---------------------------------------------------------------------------
 # WebSocket endpoint
 # ---------------------------------------------------------------------------
@@ -2655,6 +2737,77 @@ async def chat_websocket(
                 session_tasks[session_id] = asyncio.create_task(replay_demo_session(session_id))
                 continue
 
+            if msg_type == "transcribe":
+                # Voice → text: transcribe a recorded clip and return the text
+                # for the composer. No LLM turn here — the user reviews/edits
+                # the text and sends it like any typed message.
+                openai_api_key = os.environ.get("OPENAI_API_KEY", "")
+                if not openai_api_key or openai_api_key == "sk-your-key-here":
+                    await send_event(
+                        "transcription_result", request_id=request_id,
+                        success=False, error="Speech-to-text is not available (OPENAI_API_KEY not configured).",
+                    )
+                    continue
+
+                # Demo visitors are rate-limited the same way as LLM turns — a
+                # paid OpenAI call still costs money. Block before transcribing.
+                if is_demo_role(ws_user_role) and client_ip:
+                    blocked = demo_limiter.check(client_ip)
+                    if blocked:
+                        await send_event(
+                            "transcription_result", request_id=request_id,
+                            success=False, error=blocked,
+                        )
+                        continue
+
+                audio_b64 = data.get("audio_data") or ""
+                audio_format = (data.get("audio_format") or "webm").lower()
+                try:
+                    audio_bytes = base64.b64decode(audio_b64) if audio_b64 else b""
+                except Exception:
+                    audio_bytes = b""
+
+                if len(audio_bytes) < _MIN_TRANSCRIBE_BYTES:
+                    # Too short to be speech (accidental tap) — treat as empty,
+                    # no API call, no cost.
+                    await send_event(
+                        "transcription_result", request_id=request_id,
+                        success=True, text="",
+                    )
+                    continue
+
+                try:
+                    text = await asyncio.to_thread(
+                        transcribe_audio, audio_bytes, audio_format, openai_api_key,
+                    )
+                except Exception as e:
+                    logger.warning("Transcription failed: %s", e)
+                    await send_event(
+                        "transcription_result", request_id=request_id,
+                        success=False, error="Transcription failed. Please try again.",
+                    )
+                    continue
+
+                # Track the spend. Audio duration is estimated from byte size
+                # (~1.5 KB/s for webm/opus); the Demo Spend Log records every
+                # role for the admin dashboard, and counts public_manager rows
+                # against the hourly cap.
+                estimated_seconds = len(audio_bytes) / 1500
+                demo_limiter.record(
+                    client_ip or "unknown",
+                    cost_of_transcription("gpt-4o-transcribe", estimated_seconds),
+                    role=ws_user_role,
+                    provider="openai",
+                    model="gpt-4o-transcribe",
+                    session_id=data.get("session_id"),
+                )
+
+                await send_event(
+                    "transcription_result", request_id=request_id,
+                    success=True, text=text,
+                )
+                continue
+
             if msg_type != "send_message":
                 await send_error(f"Unknown message type: {msg_type}", request_id=request_id)
                 continue

{lambda_erp-0.1.17 → lambda_erp-0.1.19}/api/providers.py

@@ -43,6 +43,13 @@ ANTHROPIC_PRICING: dict[str, dict[str, Any]] = {
     "claude-haiku-4-5":  {"input": 1.00, "output": 5.00,  "cache_write": 1.25, "cache_read": 0.10},
 }
 
+# Speech-to-text, priced per minute of audio (not per token).
+TRANSCRIBE_PRICING: dict[str, dict[str, float]] = {
+    "gpt-4o-transcribe":      {"per_minute": 0.006},
+    "gpt-4o-mini-transcribe": {"per_minute": 0.003},
+    "whisper-1":              {"per_minute": 0.006},
+}
+
 
 # ---------------------------------------------------------------------------
 # Rate lookup — unknown models fall back to the most expensive model in each
@@ -114,3 +121,14 @@ def cost_of_anthropic_call(model: str, usage: Optional[Any]) -> float:
         + output_tokens * rates["output"]
     )
     return total / 1_000_000
+
+
+def cost_of_transcription(model: str, audio_seconds: float) -> float:
+    """USD cost of one speech-to-text call, billed per minute of audio.
+
+    Unknown models fall back to the most expensive STT rate so we never
+    under-count. `audio_seconds` is the (estimated) audio duration."""
+    rates = TRANSCRIBE_PRICING.get(model)
+    if rates is None:
+        rates = max(TRANSCRIBE_PRICING.values(), key=lambda r: r["per_minute"])
+    return max(float(audio_seconds or 0.0), 0.0) / 60.0 * rates["per_minute"]

{lambda_erp-0.1.17 → lambda_erp-0.1.19}/api/routers/documents.py

@@ -8,6 +8,7 @@ from api.services import (
     update_document,
     submit_document,
     cancel_document,
+    discard_document,
     convert_document,
     list_documents,
     count_documents,
@@ -29,6 +30,7 @@ def list_docs(
     from_date: str | None = None,
     to_date: str | None = None,
     docstatus: int | None = None,
+    include_discarded: bool = False,
     limit: int = Query(default=50, le=500),
     offset: int = Query(default=0, ge=0),
     _user: dict = _viewer,
@@ -44,8 +46,9 @@ def list_docs(
         filters["from_date"] = from_date
     if to_date:
         filters["to_date"] = to_date
-    rows = list_documents(doctype_slug, filters=filters, limit=limit, offset=offset)
-    total = count_documents(doctype_slug, filters=filters)
+    rows = list_documents(doctype_slug, filters=filters, limit=limit, offset=offset,
+                          include_discarded=include_discarded)
+    total = count_documents(doctype_slug, filters=filters, include_discarded=include_discarded)
     return {"rows": rows, "total": total, "limit": limit, "offset": offset}
 
 
@@ -92,6 +95,13 @@ def cancel_doc(doctype_slug: str, name: str, _user: dict = _manager):
     return cancel_document(doctype_slug, name)
 
 
+@router.post("/{doctype_slug}/{name}/discard")
+def discard_doc(doctype_slug: str, name: str, _user: dict = _manager):
+    """Void an unwanted draft (soft delete — kept for the audit trail, hidden
+    from default lists). Only valid on drafts; submitted docs must be cancelled."""
+    return discard_document(doctype_slug, name)
+
+
 @router.post("/{doctype_slug}/{name}/convert")
 def convert_doc(doctype_slug: str, name: str, data: dict, _user: dict = _manager):
     target = data.get("target_doctype")

{lambda_erp-0.1.17 → lambda_erp-0.1.19}/api/services.py

@@ -155,6 +155,16 @@ def cancel_document(doctype_slug: str, name: str) -> dict:
     return doc.as_dict()
 
 
+def discard_document(doctype_slug: str, name: str) -> dict:
+    """Void an unwanted draft (soft delete — keeps the row, hides it from lists)."""
+    doctype, cls = get_document_class(doctype_slug)
+    if not cls:
+        raise ValueError(f"Unknown document type: {doctype_slug}")
+    doc = cls.load(name)
+    doc.discard()
+    return doc.as_dict()
+
+
 def register_converter(source_doctype: str, target_doctype: str, fn) -> None:
     """Register (or override) the converter for a (source, target) pair.
 
@@ -206,7 +216,17 @@ DATE_FIELDS = {
 }
 
 
-def list_documents(doctype_slug: str, filters: dict = None, limit: int = 50, offset: int = 0) -> list:
+def _exclude_discarded(db, doctype: str, db_filters: dict, include_discarded: bool) -> None:
+    """Hide voided drafts (discarded=1) unless explicitly requested. Guarded on
+    the column existing so non-submittable doctypes are unaffected."""
+    if include_discarded:
+        return
+    if "discarded" in db._get_table_columns(doctype) and "discarded" not in db_filters:
+        db_filters["discarded"] = 0
+
+
+def list_documents(doctype_slug: str, filters: dict = None, limit: int = 50, offset: int = 0,
+                   include_discarded: bool = False) -> list:
     doctype = SLUG_TO_DOCTYPE.get(doctype_slug)
     if not doctype:
         raise ValueError(f"Unknown document type: {doctype_slug}")
@@ -226,6 +246,8 @@ def list_documents(doctype_slug: str, filters: dict = None, limit: int = 50, off
             else:
                 db_filters[key] = value
 
+    _exclude_discarded(db, doctype, db_filters, include_discarded)
+
     # Date range filtering via the doctype's primary date field
     date_field = DATE_FIELDS.get(doctype)
     if date_field:
@@ -261,7 +283,7 @@ def list_documents(doctype_slug: str, filters: dict = None, limit: int = 50, off
     return _attach_children(db, doctype_slug, rows)
 
 
-def count_documents(doctype_slug: str, filters: dict = None) -> int:
+def count_documents(doctype_slug: str, filters: dict = None, include_discarded: bool = False) -> int:
     """Count documents matching the filters (ignores limit/offset)."""
     doctype = SLUG_TO_DOCTYPE.get(doctype_slug)
     if not doctype:
@@ -282,6 +304,8 @@ def count_documents(doctype_slug: str, filters: dict = None) -> int:
             else:
                 db_filters[key] = value
 
+    _exclude_discarded(db, doctype, db_filters, include_discarded)
+
     date_field = DATE_FIELDS.get(doctype)
     where_parts = []
     params = []

{lambda_erp-0.1.17 → lambda_erp-0.1.19}/frontend/src/api/client.ts

@@ -130,6 +130,9 @@ export const api = {
   cancelDocument: (doctype: string, name: string) =>
     request<any>(`/documents/${doctype}/${encodeURIComponent(name)}/cancel`, { method: "POST" }),
 
+  discardDocument: (doctype: string, name: string) =>
+    request<any>(`/documents/${doctype}/${encodeURIComponent(name)}/discard`, { method: "POST" }),
+
   convertDocument: (doctype: string, name: string, targetDoctype: string) =>
     request<any>(`/documents/${doctype}/${encodeURIComponent(name)}/convert`, {
       method: "POST",

{lambda_erp-0.1.17 → lambda_erp-0.1.19}/lambda_erp/database.py

@@ -492,6 +492,7 @@ class Database:
                 remarks TEXT,
                 status TEXT DEFAULT 'Draft',
                 docstatus INTEGER DEFAULT 0,
+                discarded INTEGER DEFAULT 0,
                 creation TEXT,
                 modified TEXT
             )""",
@@ -545,6 +546,7 @@ class Database:
                 remarks TEXT,
                 status TEXT DEFAULT 'Draft',
                 docstatus INTEGER DEFAULT 0,
+                discarded INTEGER DEFAULT 0,
                 creation TEXT,
                 modified TEXT
             )""",
@@ -601,6 +603,7 @@ class Database:
                 remarks TEXT,
                 status TEXT DEFAULT 'Draft',
                 docstatus INTEGER DEFAULT 0,
+                discarded INTEGER DEFAULT 0,
                 creation TEXT,
                 modified TEXT
             )""",
@@ -664,6 +667,7 @@ class Database:
                 per_billed REAL DEFAULT 0,
                 status TEXT DEFAULT 'Draft',
                 docstatus INTEGER DEFAULT 0,
+                discarded INTEGER DEFAULT 0,
                 remarks TEXT,
                 creation TEXT,
                 modified TEXT
@@ -727,6 +731,7 @@ class Database:
                 purchase_order TEXT,
                 status TEXT DEFAULT 'Draft',
                 docstatus INTEGER DEFAULT 0,
+                discarded INTEGER DEFAULT 0,
                 remarks TEXT,
                 creation TEXT,
                 modified TEXT
@@ -780,6 +785,7 @@ class Database:
                 cost_center TEXT,
                 status TEXT DEFAULT 'Draft',
                 docstatus INTEGER DEFAULT 0,
+                discarded INTEGER DEFAULT 0,
                 remarks TEXT,
                 creation TEXT,
                 modified TEXT
@@ -808,6 +814,7 @@ class Database:
                 remark TEXT,
                 status TEXT DEFAULT 'Draft',
                 docstatus INTEGER DEFAULT 0,
+                discarded INTEGER DEFAULT 0,
                 creation TEXT,
                 modified TEXT
             )""",
@@ -844,6 +851,7 @@ class Database:
                 total_amount REAL DEFAULT 0,
                 status TEXT DEFAULT 'Draft',
                 docstatus INTEGER DEFAULT 0,
+                discarded INTEGER DEFAULT 0,
                 remarks TEXT,
                 creation TEXT,
                 modified TEXT
@@ -906,6 +914,7 @@ class Database:
                 return_against TEXT,
                 status TEXT DEFAULT 'Draft',
                 docstatus INTEGER DEFAULT 0,
+                discarded INTEGER DEFAULT 0,
                 remarks TEXT,
                 creation TEXT,
                 modified TEXT
@@ -958,6 +967,7 @@ class Database:
                 return_against TEXT,
                 status TEXT DEFAULT 'Draft',
                 docstatus INTEGER DEFAULT 0,
+                discarded INTEGER DEFAULT 0,
                 remarks TEXT,
                 creation TEXT,
                 modified TEXT
@@ -1015,6 +1025,7 @@ class Database:
                 return_against TEXT,
                 status TEXT DEFAULT 'Draft',
                 docstatus INTEGER DEFAULT 0,
+                discarded INTEGER DEFAULT 0,
                 remarks TEXT,
                 creation TEXT,
                 modified TEXT
@@ -1713,6 +1724,25 @@ def _m014_company_iban(db: "Database") -> None:
     db._add_column_if_missing("Company", "iban", "TEXT")
 
 
+# Submittable transactional doctypes — those with a Draft/Submitted/Cancelled
+# lifecycle. `discarded` is the soft-delete flag for voided drafts.
+_SUBMITTABLE_DOCTYPES = (
+    "Quotation", "Sales Order", "Sales Invoice", "Delivery Note",
+    "Purchase Order", "Purchase Invoice", "Purchase Receipt",
+    "Payment Entry", "Journal Entry", "Stock Entry", "POS Invoice",
+)
+
+
+def _m015_document_discarded(db: "Database") -> None:
+    """Add the `discarded` soft-delete flag to every submittable doctype.
+
+    Voiding an unwanted DRAFT sets discarded=1 (status 'Discarded') instead of
+    hard-deleting it — the row is kept for the audit trail and hidden from
+    default lists. Cancel (for submitted docs) is unaffected."""
+    for table in _SUBMITTABLE_DOCTYPES:
+        db._add_column_if_missing(table, "discarded", "INTEGER DEFAULT 0")
+
+
 Database.MIGRATIONS = [
     (1, "chat_message_session_id", _m001_chat_message_session_id),
     (2, "chat_session_user_id", _m002_chat_session_user_id),
@@ -1728,6 +1758,7 @@ Database.MIGRATIONS = [
     (12, "exchange_gain_loss_account", _m012_exchange_gain_loss_account),
     (13, "unrealized_exchange_account", _m013_unrealized_exchange_account),
     (14, "company_iban", _m014_company_iban),
+    (15, "document_discarded", _m015_document_discarded),
 ]
 
 

{lambda_erp-0.1.17 → lambda_erp-0.1.19}/lambda_erp/model.py

@@ -309,6 +309,13 @@ class Document:
             raise DocumentStatusError(
                 f"Cannot submit {self.DOCTYPE} {self.name}: docstatus is {self.docstatus}"
             )
+        # A discarded draft is voided — refuse to submit it, otherwise it would
+        # post to the ledger while staying hidden from lists ("posted but
+        # invisible"). Discard is terminal.
+        if self._data.get("discarded"):
+            raise DocumentStatusError(
+                f"Cannot submit {self.DOCTYPE} {self.name}: it was discarded."
+            )
 
         db = get_db()
         db._in_transaction = True
@@ -368,6 +375,28 @@ class Document:
         run_hooks(f"{self.DOCTYPE}:after_cancel", self)
         return self
 
+    def discard(self):
+        """Void an unwanted DRAFT (docstatus 0) — a soft delete that keeps the row.
+
+        Unlike cancel (which reverses a submission's GL/stock postings), a draft
+        has never posted anything, so there is nothing to reverse. Discarding
+        just flags the document `discarded` and sets status 'Discarded' so it
+        drops out of default lists, while the record itself is preserved for the
+        audit trail (no hard delete — see docs/agents/invariants.md). Only valid
+        on drafts; a submitted document must be cancelled instead.
+        """
+        if self.docstatus != DRAFT:
+            raise DocumentStatusError(
+                f"Cannot discard {self.DOCTYPE} {self.name}: only drafts can be "
+                f"discarded (docstatus is {self.docstatus}). A submitted document "
+                f"must be cancelled."
+            )
+        self._data["discarded"] = 1
+        self._data["status"] = "Discarded"
+        self._data["modified"] = now()
+        self._persist()
+        return self
+
     def _persist(self, commit=True):
         """Save document and child tables to database."""
         db = get_db()

{lambda_erp-0.1.17 → lambda_erp-0.1.19}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "lambda-erp"
-version = "0.1.17"
+version = "0.1.19"
 description = "Core ERP logic - accounting, sales, purchasing, inventory"
 readme = "README.md"
 license = "MIT"