feat(fusion_accounting_ocr): pluggable OCR for vendor bills

Replaces Enterprise's account_invoice_extract with a Fusion-native pipeline:

Stage 1 (text extraction): Tesseract OCRs the bill attachment via
pytesseract + pdf2image. Pluggable OCRProvider adapter pattern allows
future Mindee / Google Document AI / Ollama-vision backends.

Stage 2 (field parsing): The fusion_accounting_ai LLMProvider reads the
raw OCR text and returns structured invoice fields (vendor, invoice
number, dates, amounts, line items) as JSON.

Draft invoice fields are auto-populated for empty-only fields (never
overwriting user-entered data). Vendor matching by name against
res.partner with supplier_rank > 0.

Adds:
- account.move.ocr_state (selection: not_requested/pending/processing/
  done/failed/manual)
- account.move.ocr_raw_text, ocr_extracted_data (Json), ocr_backend,
  ocr_confidence
- fusion.ocr.log (audit trail per OCR run)
- res.company.fusion_ocr_enabled / fusion_ocr_default_backend / auto_run
- /fusion/ocr/request_for_invoice JSON-RPC endpoint

Backend availability detected at runtime via OCRProvider.is_available()
classmethods. Tesseract 5.3.4 + pytesseract 0.3.13 + pdf2image 1.17.0
are installed in the container.

Tests: 13 (TesseractAdapter availability + image OCR; flow tests for
draft autofill, no-attachment guard, customer-invoice guard, ref-not-
overwritten; field parser empty/clean-json/markdown-fence/bad-JSON/
provider-exception). All pass on westin-v19 OrbStack VM.

Made-with: Cursor

fusion_accounting_ocr/__init__.py

@@ -0,0 +1,2 @@
+from . import models
+from . import controllers

fusion_accounting_ocr/__manifest__.py

@@ -0,0 +1,39 @@
+{
+    'name': 'Fusion Accounting — Invoice OCR',
+    'version': '19.0.1.0.0',
+    'category': 'Accounting/Accounting',
+    'summary': 'OCR for vendor bills via tesseract + LLM-driven field extraction.',
+    'description': """
+Fusion Accounting — Invoice OCR
+================================
+Replaces Enterprise's account_invoice_extract with a Fusion-native pipeline:
+
+1. Tesseract OCRs the bill attachment (PDF or image) into raw text
+2. The fusion_accounting_ai LLMProvider parses the raw text into structured
+   fields (vendor, invoice number, dates, amounts, line items)
+3. Draft invoice fields are populated for the AP user to confirm
+
+Pluggable backend architecture: future Mindee, Google Document AI, or
+Ollama-vision adapters can be dropped in alongside the default tesseract
+adapter.
+""",
+    'icon': '/fusion_accounting_ocr/static/description/icon.png',
+    'author': 'Westin / Fusion Suite',
+    'depends': [
+        'fusion_accounting_core',
+        'fusion_accounting_ai',
+        'account',
+    ],
+    'external_dependencies': {
+        'python': ['pytesseract', 'pdf2image', 'PIL'],
+    },
+    'data': [
+        'security/ir.model.access.csv',
+        'views/account_move_views.xml',
+        'views/res_config_settings_views.xml',
+    ],
+    'auto_install': False,
+    'installable': True,
+    'application': False,
+    'license': 'LGPL-3',
+}

fusion_accounting_ocr/controllers/__init__.py

@@ -0,0 +1 @@
+from . import ocr_controller

fusion_accounting_ocr/controllers/ocr_controller.py

@@ -0,0 +1,21 @@
+from odoo import http
+from odoo.http import request
+
+
+class FusionOcrController(http.Controller):
+
+    @http.route('/fusion/ocr/request_for_invoice', type='jsonrpc', auth='user')
+    def request_for_invoice(self, move_id):
+        move = request.env['account.move'].browse(int(move_id))
+        move.check_access('write')
+        try:
+            move.action_request_ocr()
+            return {
+                'status': 'ok',
+                'state': move.ocr_state,
+                'backend': move.ocr_backend,
+                'confidence': move.ocr_confidence,
+                'extracted': move.ocr_extracted_data,
+            }
+        except Exception as e:
+            return {'status': 'error', 'message': str(e)}

fusion_accounting_ocr/models/__init__.py

@@ -0,0 +1,4 @@
+from . import fusion_ocr_log
+from . import res_company
+from . import res_config_settings
+from . import account_move

fusion_accounting_ocr/models/account_move.py

@@ -0,0 +1,180 @@
+"""account.move OCR extension.
+
+Adds an OCR pipeline triggered manually (or, optionally, automatically when
+a PDF/image is attached). Stage 1 is tesseract text extraction; stage 2 is
+LLM field parsing through the existing fusion_accounting_ai adapter stack.
+"""
+
+import base64
+import logging
+
+from odoo import _, fields, models
+from odoo.exceptions import UserError
+
+from ..services.ocr_providers.tesseract_adapter import TesseractAdapter
+from ..services.ocr_providers.manual_adapter import ManualAdapter
+from ..services.invoice_field_parser import parse_invoice_fields
+
+_logger = logging.getLogger(__name__)
+
+
+SUPPORTED_MIMETYPES = (
+    'application/pdf', 'image/png', 'image/jpeg', 'image/jpg',
+)
+
+
+class AccountMove(models.Model):
+    _inherit = 'account.move'
+
+    ocr_state = fields.Selection(
+        [
+            ('not_requested', 'Not Requested'),
+            ('pending', 'Pending'),
+            ('processing', 'Processing'),
+            ('done', 'Done'),
+            ('failed', 'Failed'),
+            ('manual', 'Manual Entry'),
+        ],
+        default='not_requested',
+        copy=False,
+        tracking=True,
+    )
+
+    ocr_raw_text = fields.Text(
+        string='OCR Raw Text', readonly=True, copy=False,
+        help="Raw text extracted by the OCR backend.",
+    )
+    ocr_extracted_data = fields.Json(
+        string='OCR Extracted Fields', readonly=True, copy=False,
+        help="Structured invoice fields parsed from the OCR text by the LLM.",
+    )
+    ocr_backend = fields.Char(string='OCR Backend Used', readonly=True, copy=False)
+    ocr_confidence = fields.Float(string='OCR Confidence', readonly=True, copy=False)
+    ocr_log_ids = fields.One2many('fusion.ocr.log', 'move_id', string='OCR Runs')
+
+    def action_request_ocr(self):
+        """Run OCR on the most recent supported attachment of each move."""
+        for move in self:
+            if move.move_type not in ('in_invoice', 'in_refund'):
+                raise UserError(_("OCR currently supports vendor bills only."))
+            attachment = self.env['ir.attachment'].sudo().search(
+                [
+                    ('res_model', '=', 'account.move'),
+                    ('res_id', '=', move.id),
+                    ('mimetype', 'in', SUPPORTED_MIMETYPES),
+                ],
+                order='create_date desc',
+                limit=1,
+            )
+            if not attachment:
+                raise UserError(
+                    _("No PDF or image attachment found on %s") % (move.name or move.id)
+                )
+            move._fusion_run_ocr(attachment)
+        return True
+
+    def _fusion_run_ocr(self, attachment):
+        self.ensure_one()
+        self.ocr_state = 'processing'
+
+        backend_name = (
+            self.company_id.fusion_ocr_default_backend
+            if 'fusion_ocr_default_backend' in self.company_id._fields
+            else 'tesseract'
+        )
+        provider = self._fusion_get_ocr_provider(backend_name)
+        if not provider:
+            self.ocr_state = 'manual'
+            self.message_post(
+                body=_("No OCR backend available; falling back to manual entry.")
+            )
+            return False
+
+        try:
+            data = base64.b64decode(attachment.datas)
+            result = provider.extract(
+                data, mimetype=attachment.mimetype or 'application/pdf'
+            )
+
+            self.write({
+                'ocr_raw_text': result.raw_text,
+                'ocr_backend': result.backend,
+                'ocr_confidence': result.confidence,
+            })
+            self.env['fusion.ocr.log'].sudo().create({
+                'move_id': self.id,
+                'backend': result.backend,
+                'confidence': result.confidence,
+                'raw_text_length': len(result.raw_text or ''),
+                'pages': result.pages,
+                'error': result.error,
+            })
+
+            if not result.raw_text and result.error:
+                self.ocr_state = 'failed'
+                self.message_post(body=_("OCR failed: %s") % result.error)
+                return False
+
+            parsed = parse_invoice_fields(self.env, result.raw_text)
+            self.ocr_extracted_data = parsed
+            self.ocr_state = 'done'
+
+            self._fusion_apply_ocr_fields(parsed)
+            self.message_post(
+                body=_("OCR complete: %s confidence %.0f%%") % (
+                    result.backend, (result.confidence or 0) * 100,
+                )
+            )
+            return True
+        except Exception as e:
+            _logger.exception("OCR run failed for move %s", self.id)
+            self.ocr_state = 'failed'
+            self.message_post(body=_("OCR error: %s") % e)
+            return False
+
+    def _fusion_get_ocr_provider(self, backend_name):
+        if backend_name == 'tesseract' and TesseractAdapter.is_available():
+            return TesseractAdapter()
+        if backend_name == 'manual':
+            return ManualAdapter()
+        # Future adapters (mindee, google_doc_ai, ollama_vision) plug in
+        # here. Fall back to whichever adapter is actually usable.
+        if TesseractAdapter.is_available():
+            return TesseractAdapter()
+        return ManualAdapter()
+
+    def _fusion_apply_ocr_fields(self, parsed):
+        """Apply parsed fields to a draft invoice without overwriting any
+        user-entered data. No-op on posted/cancelled invoices."""
+        if self.state != 'draft':
+            return
+
+        vals = {}
+        if parsed.get('invoice_date') and not self.invoice_date:
+            try:
+                vals['invoice_date'] = parsed['invoice_date']
+            except Exception:
+                pass
+        if parsed.get('due_date') and not self.invoice_date_due:
+            try:
+                vals['invoice_date_due'] = parsed['due_date']
+            except Exception:
+                pass
+        if parsed.get('invoice_number') and not self.ref:
+            vals['ref'] = parsed['invoice_number']
+
+        # Vendor: best-effort name match against existing supplier partners.
+        # Never auto-create a partner; AP user confirms ambiguous matches.
+        if parsed.get('vendor_name') and not self.partner_id:
+            partner = self.env['res.partner'].sudo().search(
+                [
+                    ('name', '=ilike', parsed['vendor_name']),
+                    ('supplier_rank', '>', 0),
+                ],
+                limit=1,
+            )
+            if partner:
+                vals['partner_id'] = partner.id
+
+        if vals:
+            self.write(vals)

fusion_accounting_ocr/models/fusion_ocr_log.py

@@ -0,0 +1,17 @@
+from odoo import fields, models
+
+
+class FusionOcrLog(models.Model):
+    _name = 'fusion.ocr.log'
+    _description = 'Fusion OCR Run Log'
+    _order = 'create_date desc'
+
+    move_id = fields.Many2one(
+        'account.move', required=True, ondelete='cascade', index=True,
+    )
+    backend = fields.Char(required=True)
+    confidence = fields.Float()
+    raw_text_length = fields.Integer()
+    pages = fields.Integer()
+    error = fields.Text()
+    create_date = fields.Datetime(readonly=True)

fusion_accounting_ocr/models/res_company.py

@@ -0,0 +1,26 @@
+from odoo import fields, models
+
+
+class ResCompany(models.Model):
+    _inherit = 'res.company'
+
+    fusion_ocr_enabled = fields.Boolean(
+        string='Enable Invoice OCR',
+        default=False,
+        help="When enabled, vendor bill attachments can be OCR'd via the "
+             "configured backend.",
+    )
+    fusion_ocr_default_backend = fields.Selection(
+        [
+            ('tesseract', 'Tesseract (local, free)'),
+            ('manual', 'Manual entry only'),
+        ],
+        default='tesseract',
+        string='Default OCR Backend',
+    )
+    fusion_ocr_auto_run = fields.Boolean(
+        string='Auto-run OCR on attachment',
+        default=False,
+        help="When enabled, OCR runs automatically when a PDF/image is "
+             "attached to a vendor bill.",
+    )

fusion_accounting_ocr/models/res_config_settings.py

@@ -0,0 +1,15 @@
+from odoo import fields, models
+
+
+class ResConfigSettings(models.TransientModel):
+    _inherit = 'res.config.settings'
+
+    fusion_ocr_enabled = fields.Boolean(
+        related='company_id.fusion_ocr_enabled', readonly=False,
+    )
+    fusion_ocr_default_backend = fields.Selection(
+        related='company_id.fusion_ocr_default_backend', readonly=False,
+    )
+    fusion_ocr_auto_run = fields.Boolean(
+        related='company_id.fusion_ocr_auto_run', readonly=False,
+    )

fusion_accounting_ocr/security/ir.model.access.csv

@@ -0,0 +1,3 @@
+id,name,model_id:id,group_id:id,perm_read,perm_write,perm_create,perm_unlink
+access_fusion_ocr_log_user,fusion.ocr.log.user,model_fusion_ocr_log,base.group_user,1,0,0,0
+access_fusion_ocr_log_manager,fusion.ocr.log.manager,model_fusion_ocr_log,account.group_account_manager,1,1,1,1

fusion_accounting_ocr/services/__init__.py

@@ -0,0 +1,3 @@
+from . import ocr_providers
+from . import attachment_to_image
+from . import invoice_field_parser

fusion_accounting_ocr/services/attachment_to_image.py

@@ -0,0 +1,43 @@
+"""Helper: turn an ir.attachment into a list of PIL.Image pages.
+
+Kept separate from the adapters so future backends (Ollama-vision, Mindee)
+that want PIL images directly don't have to re-implement the PDF rendering.
+"""
+
+import base64
+import io
+import logging
+
+_logger = logging.getLogger(__name__)
+
+
+def attachment_to_pages(attachment):
+    """Decode an ir.attachment into a list of PIL.Image pages.
+
+    Returns ``[]`` on failure (caller should treat as no pages).
+    """
+    try:
+        from PIL import Image
+        from pdf2image import convert_from_bytes
+    except ImportError as e:
+        _logger.warning("attachment_to_pages requires PIL + pdf2image: %s", e)
+        return []
+
+    if not attachment or not attachment.datas:
+        return []
+
+    try:
+        data = base64.b64decode(attachment.datas)
+    except Exception as e:
+        _logger.warning("Could not decode attachment %s: %s", attachment.id, e)
+        return []
+
+    mimetype = attachment.mimetype or ''
+    is_pdf = mimetype == 'application/pdf' or data[:4] == b'%PDF'
+    try:
+        if is_pdf:
+            return convert_from_bytes(data, dpi=200)
+        return [Image.open(io.BytesIO(data))]
+    except Exception as e:
+        _logger.warning("Could not render attachment %s: %s", attachment.id, e)
+        return []

fusion_accounting_ocr/services/invoice_field_parser.py

@@ -0,0 +1,150 @@
+"""Stage-2 of the OCR pipeline: parse raw OCR text into structured invoice
+fields via the configured LLM provider.
+
+Mirrors the pattern in fusion_accounting_followup/services/followup_text_generator.py:
+look up an adapter by ir.config_parameter, fall back gracefully when no
+provider is configured, and never let an LLM hiccup nuke the OCR result.
+"""
+
+import json
+import logging
+
+_logger = logging.getLogger(__name__)
+
+
+SYSTEM_PROMPT = (
+    "You are an invoice field extraction assistant. You read raw OCR text "
+    "from vendor bills and return a strict JSON object with the requested "
+    "fields. You never include commentary or markdown fences. When a field "
+    "cannot be determined from the text you return null for that field."
+)
+
+USER_PROMPT = """Given the raw OCR text of a vendor bill, return a JSON object
+with these fields (use null when unclear):
+
+{{
+  "vendor_name": <string, the seller/vendor company name>,
+  "invoice_number": <string, the bill or invoice reference number>,
+  "invoice_date": <string, ISO format YYYY-MM-DD>,
+  "due_date": <string or null, ISO format YYYY-MM-DD>,
+  "currency": <string, ISO 4217 code like CAD/USD/EUR>,
+  "subtotal": <number or null>,
+  "tax_total": <number or null>,
+  "total": <number, the grand total amount due>,
+  "line_items": [
+    {{"description": <string>, "quantity": <number or null>,
+      "unit_price": <number or null>, "amount": <number or null>}}
+  ]
+}}
+
+Return ONLY valid JSON, no commentary, no markdown fences.
+
+Raw OCR text:
+---
+{text}
+---
+"""
+
+
+def parse_invoice_fields(env, raw_text: str, *, provider=None) -> dict:
+    """Use the configured LLM provider to extract structured invoice fields.
+
+    Returns a dict with the schema above. On any failure (no provider, bad
+    JSON, network error, etc.) returns an all-null result so the OCR raw
+    text is still preserved for the AP user.
+    """
+    if not raw_text or not raw_text.strip():
+        return _empty_result()
+
+    if provider is None:
+        provider = _get_provider(env)
+    if provider is None:
+        _logger.info(
+            "No LLM provider configured for OCR field parsing; "
+            "raw OCR text preserved, fields left empty."
+        )
+        return _empty_result()
+
+    try:
+        truncated = raw_text[:12000]
+        user = USER_PROMPT.format(text=truncated)
+        response = provider.complete(
+            system=SYSTEM_PROMPT,
+            messages=[{'role': 'user', 'content': user}],
+            max_tokens=1000,
+            temperature=0.1,
+        )
+        content = response.get('content') if isinstance(response, dict) else response
+        if not content:
+            return _empty_result()
+
+        # LLMs sometimes wrap JSON in ```json ... ``` despite instructions.
+        content = content.strip()
+        if content.startswith('```'):
+            content = content.split('```', 2)[1]
+            if content.startswith('json'):
+                content = content[4:]
+            content = content.rsplit('```', 1)[0]
+
+        parsed = json.loads(content.strip())
+        return {
+            'vendor_name': parsed.get('vendor_name'),
+            'invoice_number': parsed.get('invoice_number'),
+            'invoice_date': parsed.get('invoice_date'),
+            'due_date': parsed.get('due_date'),
+            'currency': parsed.get('currency'),
+            'subtotal': parsed.get('subtotal'),
+            'tax_total': parsed.get('tax_total'),
+            'total': parsed.get('total'),
+            'line_items': parsed.get('line_items') or [],
+        }
+    except json.JSONDecodeError as e:
+        _logger.warning("LLM returned non-JSON for OCR field parsing: %s", e)
+        return _empty_result()
+    except Exception as e:
+        _logger.warning("OCR field parsing failed: %s", e)
+        return _empty_result()
+
+
+def _empty_result():
+    return {
+        'vendor_name': None,
+        'invoice_number': None,
+        'invoice_date': None,
+        'due_date': None,
+        'currency': None,
+        'subtotal': None,
+        'tax_total': None,
+        'total': None,
+        'line_items': [],
+    }
+
+
+def _get_provider(env):
+    """Look up the LLM adapter via ir.config_parameter.
+
+    Honours a feature-specific override
+    (``fusion_accounting.provider.ocr_field_parsing``) before falling back
+    to the suite-wide default (``fusion_accounting.provider.default``).
+    Returns None when no adapter is configured/importable.
+    """
+    param = env['ir.config_parameter'].sudo()
+    name = param.get_param('fusion_accounting.provider.ocr_field_parsing')
+    if not name:
+        name = param.get_param('fusion_accounting.provider.default')
+    if not name:
+        return None
+    try:
+        from odoo.addons.fusion_accounting_ai.services.adapters.openai_adapter import OpenAIAdapter
+        from odoo.addons.fusion_accounting_ai.services.adapters.claude import ClaudeAdapter
+    except ImportError:
+        return None
+    try:
+        if name.startswith('openai'):
+            return OpenAIAdapter(env)
+        if name.startswith('claude'):
+            return ClaudeAdapter(env)
+    except Exception as e:
+        _logger.warning("OCR field parser could not instantiate %s: %s", name, e)
+        return None
+    return None

fusion_accounting_ocr/services/ocr_providers/__init__.py

@@ -0,0 +1,3 @@
+from . import base
+from . import tesseract_adapter
+from . import manual_adapter

fusion_accounting_ocr/services/ocr_providers/base.py

@@ -0,0 +1,40 @@
+"""OCRProvider contract - every backend must conform.
+
+Mirrors the LLMProvider pattern in fusion_accounting_ai. Future adapters
+(Mindee, Google Document AI, Ollama-vision) drop in alongside the default
+tesseract adapter without touching account.move.
+"""
+
+from abc import ABC, abstractmethod
+from dataclasses import dataclass, field
+
+
+@dataclass
+class OCRResult:
+    raw_text: str = ''
+    confidence: float = 0.0  # 0.0–1.0
+    pages: int = 0
+    backend: str = ''
+    error: str = ''
+    metadata: dict = field(default_factory=dict)
+
+
+class OCRProvider(ABC):
+    """Abstract OCR backend. Subclasses implement extract()."""
+
+    name: str = 'base'
+
+    @abstractmethod
+    def extract(self, image_or_pdf_bytes: bytes, *, mimetype: str = 'application/pdf') -> OCRResult:
+        """Extract text from raw bytes.
+
+        ``mimetype`` hints whether to PDF-render (poppler) or image-decode
+        (PIL) the bytes. Implementations should still inspect the byte
+        signature for safety.
+        """
+        ...
+
+    @classmethod
+    def is_available(cls) -> bool:
+        """Return True if the backend's runtime deps are present."""
+        return True

fusion_accounting_ocr/services/ocr_providers/manual_adapter.py

@@ -0,0 +1,13 @@
+"""Manual fallback adapter - no real OCR, just marks the document as
+'awaiting manual entry'. Used when no real OCR backend is available
+or when the user explicitly disables OCR.
+"""
+
+from .base import OCRProvider, OCRResult
+
+
+class ManualAdapter(OCRProvider):
+    name = 'manual'
+
+    def extract(self, image_or_pdf_bytes, *, mimetype='application/pdf'):
+        return OCRResult(raw_text='', confidence=0.0, pages=0, backend='manual')

fusion_accounting_ocr/services/ocr_providers/tesseract_adapter.py

@@ -0,0 +1,71 @@
+"""Tesseract OCR adapter.
+
+Uses the system tesseract binary via pytesseract, with poppler-backed
+PDF rendering via pdf2image. Inside the container these are pre-installed:
+- tesseract-ocr 5.3.4
+- pytesseract 0.3.13
+- pdf2image 1.17.0
+- poppler-utils
+"""
+
+import io
+import logging
+
+from .base import OCRProvider, OCRResult
+
+_logger = logging.getLogger(__name__)
+
+
+class TesseractAdapter(OCRProvider):
+    name = 'tesseract'
+
+    @classmethod
+    def is_available(cls) -> bool:
+        try:
+            import pytesseract
+            from pdf2image import convert_from_bytes  # noqa: F401
+            from PIL import Image  # noqa: F401
+            pytesseract.get_tesseract_version()
+            return True
+        except Exception as e:
+            _logger.debug("TesseractAdapter not available: %s", e)
+            return False
+
+    def extract(self, image_or_pdf_bytes, *, mimetype='application/pdf'):
+        import pytesseract
+        from pdf2image import convert_from_bytes
+        from PIL import Image
+
+        try:
+            is_pdf = (
+                mimetype == 'application/pdf'
+                or (image_or_pdf_bytes[:4] == b'%PDF')
+            )
+            if is_pdf:
+                pages = convert_from_bytes(image_or_pdf_bytes, dpi=200)
+            else:
+                img = Image.open(io.BytesIO(image_or_pdf_bytes))
+                pages = [img]
+
+            texts = []
+            for p in pages:
+                texts.append(pytesseract.image_to_string(p))
+            full_text = '\n\f\n'.join(texts)
+
+            # Heuristic confidence - tesseract has a per-word conf in
+            # image_to_data, but a length proxy is fine for routing
+            # decisions. Future: use pytesseract.image_to_data for a real
+            # average word-level confidence.
+            conf = min(1.0, len(full_text) / 1000.0)
+            return OCRResult(
+                raw_text=full_text,
+                confidence=conf,
+                pages=len(pages),
+                backend='tesseract',
+            )
+        except Exception as e:
+            _logger.warning("Tesseract OCR failed: %s", e)
+            return OCRResult(
+                raw_text='', confidence=0.0, pages=0,
+                backend='tesseract', error=str(e),
+            )

fusion_accounting_ocr/tests/__init__.py

@@ -0,0 +1,3 @@
+from . import test_tesseract_adapter
+from . import test_invoice_ocr_flow
+from . import test_field_parser

fusion_accounting_ocr/tests/test_field_parser.py

@@ -0,0 +1,74 @@
+from unittest.mock import MagicMock
+
+from odoo.tests import tagged
+from odoo.tests.common import TransactionCase
+
+from odoo.addons.fusion_accounting_ocr.services.invoice_field_parser import (
+    parse_invoice_fields,
+)
+
+
+@tagged('post_install', '-at_install')
+class TestFieldParser(TransactionCase):
+
+    def test_parser_handles_empty_text(self):
+        result = parse_invoice_fields(self.env, '')
+        self.assertIsNone(result['total'])
+        self.assertEqual(result['line_items'], [])
+
+    def test_parser_handles_no_provider_gracefully(self):
+        # Without an LLM provider configured, parse should return an empty
+        # result dict rather than crashing.
+        result = parse_invoice_fields(self.env, 'INVOICE 12345 Total $100')
+        self.assertIn('total', result)
+        self.assertIn('line_items', result)
+        self.assertIsInstance(result['line_items'], list)
+
+    def test_parser_consumes_clean_json(self):
+        provider = MagicMock()
+        provider.complete.return_value = {
+            'content': (
+                '{"vendor_name": "Acme Co", "invoice_number": "INV-1",'
+                ' "invoice_date": "2026-04-20", "due_date": null,'
+                ' "currency": "CAD", "subtotal": 90.0, "tax_total": 10.0,'
+                ' "total": 100.0, "line_items": ['
+                '{"description": "Widget", "quantity": 1, "unit_price": 90.0,'
+                ' "amount": 90.0}]}'
+            ),
+        }
+        result = parse_invoice_fields(self.env, 'raw text', provider=provider)
+        self.assertEqual(result['vendor_name'], 'Acme Co')
+        self.assertEqual(result['invoice_number'], 'INV-1')
+        self.assertEqual(result['total'], 100.0)
+        self.assertEqual(len(result['line_items']), 1)
+        self.assertEqual(result['line_items'][0]['description'], 'Widget')
+
+    def test_parser_strips_markdown_fences(self):
+        provider = MagicMock()
+        provider.complete.return_value = {
+            'content': (
+                '```json\n'
+                '{"vendor_name": "Beta Ltd", "invoice_number": "B-2",'
+                ' "invoice_date": null, "due_date": null, "currency": null,'
+                ' "subtotal": null, "tax_total": null, "total": 5.5,'
+                ' "line_items": []}\n'
+                '```'
+            ),
+        }
+        result = parse_invoice_fields(self.env, 'raw text', provider=provider)
+        self.assertEqual(result['vendor_name'], 'Beta Ltd')
+        self.assertEqual(result['total'], 5.5)
+
+    def test_parser_returns_empty_on_invalid_json(self):
+        provider = MagicMock()
+        provider.complete.return_value = {'content': 'not json at all'}
+        result = parse_invoice_fields(self.env, 'raw text', provider=provider)
+        self.assertIsNone(result['total'])
+        self.assertEqual(result['line_items'], [])
+
+    def test_parser_returns_empty_on_provider_exception(self):
+        provider = MagicMock()
+        provider.complete.side_effect = RuntimeError('boom')
+        result = parse_invoice_fields(self.env, 'raw text', provider=provider)
+        self.assertIsNone(result['total'])
+        self.assertEqual(result['line_items'], [])

fusion_accounting_ocr/tests/test_invoice_ocr_flow.py

@@ -0,0 +1,117 @@
+import base64
+import io
+from unittest.mock import patch
+
+from PIL import Image, ImageDraw
+
+from odoo.exceptions import UserError
+from odoo.tests import tagged
+from odoo.tests.common import TransactionCase
+
+
+@tagged('post_install', '-at_install')
+class TestInvoiceOcrFlow(TransactionCase):
+
+    def setUp(self):
+        super().setUp()
+        self.partner = self.env['res.partner'].create({
+            'name': 'Test Vendor',
+            'supplier_rank': 1,
+        })
+        self.move = self.env['account.move'].create({
+            'move_type': 'in_invoice',
+            'partner_id': self.partner.id,
+        })
+
+    def test_ocr_state_default(self):
+        self.assertEqual(self.move.ocr_state, 'not_requested')
+
+    def test_action_request_ocr_no_attachment_raises(self):
+        with self.assertRaises(UserError):
+            self.move.action_request_ocr()
+
+    def test_action_request_ocr_with_image(self):
+        img = Image.new('RGB', (800, 120), color='white')
+        draw = ImageDraw.Draw(img)
+        try:
+            from PIL import ImageFont
+            font = ImageFont.truetype(
+                '/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf', 36,
+            )
+        except Exception:
+            font = None
+        draw.text((20, 30), "TOTAL $50.00 INV-9999", fill='black', font=font)
+        buf = io.BytesIO()
+        img.save(buf, format='PNG')
+
+        self.env['ir.attachment'].create({
+            'name': 'test_invoice.png',
+            'datas': base64.b64encode(buf.getvalue()),
+            'res_model': 'account.move',
+            'res_id': self.move.id,
+            'mimetype': 'image/png',
+        })
+
+        # Mock the LLM call to avoid a real API roundtrip.
+        with patch(
+            'odoo.addons.fusion_accounting_ocr.models.account_move.parse_invoice_fields',
+            return_value={
+                'vendor_name': None,
+                'invoice_number': 'INV-9999',
+                'invoice_date': None,
+                'due_date': None,
+                'currency': None,
+                'subtotal': None,
+                'tax_total': None,
+                'total': 50.0,
+                'line_items': [],
+            },
+        ):
+            self.move.action_request_ocr()
+
+        self.assertEqual(self.move.ocr_state, 'done')
+        self.assertEqual(self.move.ocr_backend, 'tesseract')
+        self.assertGreater(self.move.ocr_confidence, 0)
+        self.assertIsNotNone(self.move.ocr_extracted_data)
+        # Parsed invoice_number should land on the invoice's ref field.
+        self.assertEqual(self.move.ref, 'INV-9999')
+        # OCR log row was created.
+        self.assertEqual(len(self.move.ocr_log_ids), 1)
+        log = self.move.ocr_log_ids
+        self.assertEqual(log.backend, 'tesseract')
+        self.assertGreater(log.raw_text_length, 0)
+
+    def test_apply_does_not_overwrite_user_entered_ref(self):
+        self.move.ref = 'USER-SET-REF'
+        img = Image.new('RGB', (400, 80), color='white')
+        ImageDraw.Draw(img).text((10, 30), "INV-7777", fill='black')
+        buf = io.BytesIO()
+        img.save(buf, format='PNG')
+        self.env['ir.attachment'].create({
+            'name': 't.png',
+            'datas': base64.b64encode(buf.getvalue()),
+            'res_model': 'account.move',
+            'res_id': self.move.id,
+            'mimetype': 'image/png',
+        })
+        with patch(
+            'odoo.addons.fusion_accounting_ocr.models.account_move.parse_invoice_fields',
+            return_value={
+                'vendor_name': None, 'invoice_number': 'INV-7777',
+                'invoice_date': None, 'due_date': None, 'currency': None,
+                'subtotal': None, 'tax_total': None, 'total': None,
+                'line_items': [],
+            },
+        ):
+            self.move.action_request_ocr()
+
+        # User-entered ref must not be overwritten.
+        self.assertEqual(self.move.ref, 'USER-SET-REF')
+
+    def test_only_vendor_bills_supported(self):
+        customer_invoice = self.env['account.move'].create({
+            'move_type': 'out_invoice',
+            'partner_id': self.partner.id,
+        })
+        with self.assertRaises(UserError):
+            customer_invoice.action_request_ocr()

fusion_accounting_ocr/tests/test_tesseract_adapter.py

@@ -0,0 +1,47 @@
+import io
+
+from PIL import Image, ImageDraw
+
+from odoo.tests import tagged
+from odoo.tests.common import TransactionCase
+
+from odoo.addons.fusion_accounting_ocr.services.ocr_providers.tesseract_adapter import (
+    TesseractAdapter,
+)
+
+
+@tagged('post_install', '-at_install')
+class TestTesseractAdapter(TransactionCase):
+
+    def test_is_available(self):
+        # In our container tesseract + pytesseract + pdf2image are pre-installed.
+        self.assertTrue(TesseractAdapter.is_available())
+
+    def test_extract_simple_text_image(self):
+        # Generate a tiny PNG with the text "INVOICE 12345 Total $100".
+        # Use a slightly larger image and try to load a TTF font for
+        # tesseract reliability; fall back to default bitmap font otherwise.
+        img = Image.new('RGB', (800, 120), color='white')
+        draw = ImageDraw.Draw(img)
+        try:
+            from PIL import ImageFont
+            font = ImageFont.truetype(
+                '/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf', 36,
+            )
+        except Exception:
+            font = None
+        draw.text((20, 30), "INVOICE 12345 Total $100", fill='black', font=font)
+
+        buf = io.BytesIO()
+        img.save(buf, format='PNG')
+        png_bytes = buf.getvalue()
+
+        adapter = TesseractAdapter()
+        result = adapter.extract(png_bytes, mimetype='image/png')
+
+        self.assertEqual(result.backend, 'tesseract')
+        self.assertEqual(result.error, '')
+        self.assertEqual(result.pages, 1)
+        self.assertGreater(len(result.raw_text), 0)
+        # Tesseract should pick up the digits at minimum.
+        self.assertIn('12345', result.raw_text.replace(' ', ''))

fusion_accounting_ocr/views/account_move_views.xml

@@ -0,0 +1,45 @@
+<?xml version="1.0" encoding="utf-8"?>
+<odoo>
+
+    <record id="view_move_form_inherit_fusion_ocr" model="ir.ui.view">
+        <field name="name">account.move.form.inherit.fusion_ocr</field>
+        <field name="model">account.move</field>
+        <field name="inherit_id" ref="account.view_move_form"/>
+        <field name="arch" type="xml">
+
+            <xpath expr="//header" position="inside">
+                <button name="action_request_ocr"
+                        type="object"
+                        string="Request OCR"
+                        class="oe_highlight"
+                        invisible="move_type not in ('in_invoice', 'in_refund') or ocr_state in ('processing', 'done')"/>
+                <button name="action_request_ocr"
+                        type="object"
+                        string="Re-run OCR"
+                        invisible="move_type not in ('in_invoice', 'in_refund') or ocr_state not in ('done', 'failed', 'manual')"/>
+            </xpath>
+
+            <xpath expr="//sheet" position="inside">
+                <group string="Fusion OCR"
+                       invisible="move_type not in ('in_invoice', 'in_refund') or ocr_state == 'not_requested'">
+                    <group>
+                        <field name="ocr_state" widget="badge"
+                               decoration-success="ocr_state == 'done'"
+                               decoration-info="ocr_state == 'processing'"
+                               decoration-warning="ocr_state == 'manual'"
+                               decoration-danger="ocr_state == 'failed'"/>
+                        <field name="ocr_backend" readonly="1"/>
+                        <field name="ocr_confidence" readonly="1" widget="percentage"/>
+                    </group>
+                    <group>
+                        <field name="ocr_extracted_data" readonly="1" widget="text"/>
+                    </group>
+                    <field name="ocr_raw_text" readonly="1" nolabel="1"
+                           placeholder="Raw OCR text..."/>
+                </group>
+            </xpath>
+
+        </field>
+    </record>
+
+</odoo>

fusion_accounting_ocr/views/res_config_settings_views.xml

@@ -0,0 +1,35 @@
+<?xml version="1.0" encoding="utf-8"?>
+<odoo>
+
+    <record id="res_config_settings_view_form_inherit_fusion_ocr" model="ir.ui.view">
+        <field name="name">res.config.settings.view.form.inherit.fusion_ocr</field>
+        <field name="model">res.config.settings</field>
+        <field name="inherit_id" ref="account.res_config_settings_view_form"/>
+        <field name="arch" type="xml">
+
+            <xpath expr="//block[@id='account_vendor_bills']" position="after">
+                <block title="Fusion Invoice OCR" id="fusion_ocr_section">
+                    <setting id="fusion_ocr_enabled_setting"
+                             string="Enable Invoice OCR"
+                             help="OCR vendor bill attachments via the configured backend.">
+                        <field name="fusion_ocr_enabled"/>
+                        <div class="content-group" invisible="not fusion_ocr_enabled">
+                            <div class="mt16">
+                                <label for="fusion_ocr_default_backend"
+                                       string="Default OCR Backend" class="o_light_label"/>
+                                <field name="fusion_ocr_default_backend"/>
+                            </div>
+                            <div class="mt16">
+                                <field name="fusion_ocr_auto_run"/>
+                                <label for="fusion_ocr_auto_run"
+                                       string="Auto-run OCR on attachment"/>
+                            </div>
+                        </div>
+                    </setting>
+                </block>
+            </xpath>
+
+        </field>
+    </record>
+
+</odoo>