2453bc6ef2
- Ollama container running llama3.2-vision:11b on server - OCR page in ops app: camera/upload → Ollama extracts vendor, date, amounts, line items → editable form → create Purchase Invoice - nginx proxies /ollama/ to Ollama API (both ops + field containers) - Added createDoc to erp.js API layer Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
84 lines
2.5 KiB
JavaScript
84 lines
2.5 KiB
JavaScript
import { authFetch } from './auth'
|
|
|
|
const OLLAMA_URL = '/ollama/api/generate'
|
|
|
|
const OCR_PROMPT = `You are an invoice/bill OCR assistant. Extract the following fields from this image of a bill or invoice. Return ONLY valid JSON, no markdown, no explanation.
|
|
|
|
{
|
|
"vendor": "company name on the bill",
|
|
"vendor_address": "full address if visible",
|
|
"invoice_number": "invoice/bill number",
|
|
"date": "YYYY-MM-DD format",
|
|
"due_date": "YYYY-MM-DD if visible, null otherwise",
|
|
"subtotal": 0.00,
|
|
"tax_gst": 0.00,
|
|
"tax_qst": 0.00,
|
|
"total": 0.00,
|
|
"currency": "CAD",
|
|
"items": [
|
|
{ "description": "line item description", "qty": 1, "rate": 0.00, "amount": 0.00 }
|
|
],
|
|
"notes": "any other relevant text (account number, payment terms, etc.)"
|
|
}
|
|
|
|
If a field is not visible, set it to null. Always return valid JSON.`
|
|
|
|
/**
|
|
* Send an image to Ollama Vision for bill/invoice OCR.
|
|
* @param {string} base64Image — base64 encoded image (no data: prefix)
|
|
* @returns {object} Parsed invoice data
|
|
*/
|
|
export async function ocrBill (base64Image) {
|
|
// Strip data:image/...;base64, prefix if present
|
|
const clean = base64Image.replace(/^data:image\/[^;]+;base64,/, '')
|
|
|
|
const res = await authFetch(OLLAMA_URL, {
|
|
method: 'POST',
|
|
headers: { 'Content-Type': 'application/json' },
|
|
body: JSON.stringify({
|
|
model: 'llama3.2-vision:11b',
|
|
prompt: OCR_PROMPT,
|
|
images: [clean],
|
|
stream: false,
|
|
options: {
|
|
temperature: 0.1,
|
|
num_predict: 2048,
|
|
},
|
|
}),
|
|
})
|
|
|
|
if (!res.ok) {
|
|
const text = await res.text()
|
|
throw new Error('OCR failed: ' + (text || res.status))
|
|
}
|
|
|
|
const data = await res.json()
|
|
const raw = data.response || ''
|
|
|
|
// Extract JSON from response (model might wrap it in markdown)
|
|
const jsonMatch = raw.match(/\{[\s\S]*\}/)
|
|
if (!jsonMatch) throw new Error('No JSON in OCR response')
|
|
|
|
try {
|
|
return JSON.parse(jsonMatch[0])
|
|
} catch (e) {
|
|
throw new Error('Invalid JSON from OCR: ' + e.message)
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Check if Ollama is running and the vision model is available.
|
|
*/
|
|
export async function checkOllamaStatus () {
|
|
try {
|
|
const res = await authFetch('/ollama/api/tags')
|
|
if (!res.ok) return { online: false, error: 'HTTP ' + res.status }
|
|
const data = await res.json()
|
|
const models = (data.models || []).map(m => m.name)
|
|
const hasVision = models.some(m => m.includes('llama3.2-vision'))
|
|
return { online: true, models, hasVision }
|
|
} catch (e) {
|
|
return { online: false, error: e.message }
|
|
}
|
|
}
|