From 3327fd4face2059b543a5d727734d6a2b7883625 Mon Sep 17 00:00:00 2001
From: cclohmar <claus.lohmar@lohmar.co.uk>
Date: Sat, 30 May 2026 17:17:06 +0000
Subject: [PATCH] feat: use glm-ocr as default Ollama model (specialized OCR,
 not a general LLM)

- glm-ocr is a 1.1B parameter model built specifically for OCR
- No reasoning overhead, no thinking field issues
- Faster inference than qwen3.5 on CPU
- Removed old qwen3.5 models (2B + 0.8B) to free ~4GB disk
- Updated install.sh, ollama.go, .env.example defaults
---
 .env.example          | 2 +-
 install.sh            | 2 +-
 internal/ai/ollama.go | 2 +-
 3 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/.env.example b/.env.example
index 9f10a6f..eec0064 100644
--- a/.env.example
+++ b/.env.example
@@ -16,7 +16,7 @@ AI_PROVIDER=gemini
 
 # For AI_PROVIDER=ollama:
 # AI_BASE_URL=http://localhost:11434
-# AI_MODEL=qwen3.5:2b
+# AI_MODEL=glm-ocr
 
 # --- SMTP (optional — needed for OTP emails and report delivery) ---
 # SMTP_HOST=smtp.example.com
diff --git a/install.sh b/install.sh
index 094e13d..45b68c1 100755
--- a/install.sh
+++ b/install.sh
@@ -176,7 +176,7 @@ case "$AI_CHOICE" in
         ;;
     3)
         AI_PROVIDER="ollama"
-        AI_MODEL="qwen3.5:2b"
+        AI_MODEL="glm-ocr"
         # Ollama's install script requires zstd
         if ! command -v zstd &>/dev/null; then
             info "Installing zstd (required by Ollama)..."
diff --git a/internal/ai/ollama.go b/internal/ai/ollama.go
index c4b28fa..df15151 100644
--- a/internal/ai/ollama.go
+++ b/internal/ai/ollama.go
@@ -67,7 +67,7 @@ func (p *ollamaProvider) ExtractReceipt(imagePath string) (*ReceiptData, error)
 
 	model := os.Getenv("AI_MODEL")
 	if model == "" {
-		model = "qwen3.5:2b"
+		model = "glm-ocr"
 	}
 
 	b64Data := base64.StdEncoding.EncodeToString(imageData)