fix: Render standard-font PDFs in the proxy image, add a Gemini PDF example

CHANGELOG.md

@@ -3,6 +3,12 @@
 Releases are cut by pushing a `v*` tag, which publishes the images to Docker Hub.
 Entries before v0.4.0 were reconstructed from git history.
 
+## Unreleased
+
+### Fixed
+
+- The `proxy` image now renders PDFs that use the standard PDF fonts (Helvetica, Times, Courier). Alpine's `poppler-utils` ships without them, so those pages previously rasterized blank and the model saw nothing. Adds `font-liberation` (about 5 MB, so the image is now ~46 MB).
+
 ## v0.4.1 (2026-08-06)
 
 ### Changed

Dockerfile

@@ -5,7 +5,9 @@ COPY . .
 RUN CGO_ENABLED=0 GOOS=$TARGETOS GOARCH=$TARGETARCH go build -ldflags="-s -w" -o /out/localaik ./cmd/localaik
 
 FROM alpine:3@sha256:28bd5fe8b56d1bd048e5babf5b10710ebe0bae67db86916198a6eec434943f8b AS proxy
-RUN apk add --no-cache ca-certificates poppler-utils tini
+# font-liberation gives poppler the Base-14 fonts (Helvetica, Times, Courier);
+# without it, standard-font PDFs render as blank pages.
+RUN apk add --no-cache ca-certificates poppler-utils font-liberation tini
 COPY --from=proxy-builder /out/localaik /usr/local/bin/localaik
 ENV PORT=8090
 # No inference engine here, so the loopback default could never work.

README.md

@@ -135,7 +135,7 @@ client := anthropic.NewClient(
 | --------------------- | ------------------ | ---------- |
 | `latest`, `gemma3-4b` | Gemma 3 4B Q4_K_M  | ~3 GB      |
 | `gemma3-12b`          | Gemma 3 12B Q4_K_M | ~7 GB      |
-| `proxy`               | none (you supply)  | ~41 MB     |
+| `proxy`               | none (you supply)  | ~46 MB     |
 
 
 Version-pinned tags follow the pattern `v0.1.1-gemma3-4b`, `v0.1.1-gemma3-12b`,

examples/README.md

@@ -10,7 +10,7 @@ Small, runnable samples that talk to **localaik** on `http://localhost:8090`. Us
    docker run -d -p 8090:8090 gokhalh/localaik
    ```
 
-   Or from the repo root: `make docker-up` (defaults to port `18090` — set `PORT=8090` if you want the examples unchanged).
+   Or from the repo root: `make docker-up` (defaults to port `18090`; set `PORT=8090` if you want the examples unchanged).
 
 2. Wait until the model is loaded (`GET /health` returns 200). The first start can take a while.
 
@@ -31,6 +31,8 @@ Small, runnable samples that talk to **localaik** on `http://localhost:8090`. Us
 | **JavaScript** | [javascript/gemini](javascript/gemini/index.mjs) | [javascript/openai](javascript/openai/index.mjs) | not yet | [javascript/gemini-structured](javascript/gemini-structured/index.mjs) |
 | **Java**    | [java/gemini](java/gemini/Gemini.java) | [java/openai](java/openai/OpenAI.java) | not yet | [java/gemini-structured](java/gemini-structured/GeminiStructured.java) |
 
+**PDF parsing:** [go/gemini-pdf](go/gemini-pdf/main.go) sends a fake invoice PDF through Gemini and extracts its fields as structured JSON. It ships an embedded sample, so `go run main.go` works with no setup; pass a path to use your own PDF.
+
 ## Conventions
 
 - **Base URL:** `http://localhost:8090` for Gemini-style calls; OpenAI clients use `http://localhost:8090/v1`; Anthropic clients use `http://localhost:8090` (their SDKs append `v1/` themselves).

examples/go/gemini-pdf/main.go

@@ -0,0 +1,90 @@
+package main
+
+import (
+	"context"
+	_ "embed"
+	"encoding/json"
+	"fmt"
+	"log"
+	"os"
+
+	"google.golang.org/genai"
+)
+
+// A small fake invoice so the example runs with no setup. Pass a path argument
+// to use your own PDF instead.
+//
+//go:embed sample.pdf
+var samplePDF []byte
+
+type invoice struct {
+	InvoiceNumber string `json:"invoice_number"`
+	Vendor        string `json:"vendor"`
+	AmountDue     string `json:"amount_due"`
+	DueDate       string `json:"due_date"`
+}
+
+func main() {
+	ctx := context.Background()
+
+	pdfBytes := samplePDF
+	if len(os.Args) > 1 {
+		data, err := os.ReadFile(os.Args[1])
+		if err != nil {
+			log.Fatal(err)
+		}
+		pdfBytes = data
+	}
+
+	client, err := genai.NewClient(ctx, &genai.ClientConfig{
+		APIKey:  "test",
+		Backend: genai.BackendGeminiAPI,
+		HTTPOptions: genai.HTTPOptions{
+			BaseURL: "http://localhost:8090",
+		},
+	})
+	if err != nil {
+		log.Fatal(err)
+	}
+
+	// A schema plus temperature 0 make the model read the fields off the page
+	// rather than inventing plausible-looking ones.
+	config := &genai.GenerateContentConfig{
+		Temperature:      genai.Ptr[float32](0),
+		ResponseMIMEType: "application/json",
+		ResponseSchema: &genai.Schema{
+			Type: genai.TypeObject,
+			Properties: map[string]*genai.Schema{
+				"invoice_number": {Type: genai.TypeString},
+				"vendor":         {Type: genai.TypeString},
+				"amount_due":     {Type: genai.TypeString},
+				"due_date":       {Type: genai.TypeString},
+			},
+			Required: []string{"invoice_number", "vendor", "amount_due", "due_date"},
+		},
+	}
+
+	resp, err := client.Models.GenerateContent(ctx,
+		"localaik",
+		[]*genai.Content{{
+			Parts: []*genai.Part{
+				{Text: "Extract the invoice fields from this document."},
+				genai.NewPartFromBytes(pdfBytes, "application/pdf"),
+			},
+		}},
+		config,
+	)
+	if err != nil {
+		log.Fatal(err)
+	}
+
+	var inv invoice
+	if err := json.Unmarshal([]byte(resp.Text()), &inv); err != nil {
+		log.Fatalf("response was not valid JSON: %v\nraw: %s", err, resp.Text())
+	}
+
+	fmt.Printf("invoice number: %s\n", inv.InvoiceNumber)
+	fmt.Printf("vendor:         %s\n", inv.Vendor)
+	fmt.Printf("amount due:     %s\n", inv.AmountDue)
+	fmt.Printf("due date:       %s\n", inv.DueDate)
+}

integration/proxy_image_test.go

@@ -170,12 +170,28 @@ func TestProxyImageRoundTripsAllProtocols(t *testing.T) {
 		if err != nil {
 			t.Fatalf("rendered page was not valid base64: %v", err)
 		}
-		config, err := png.DecodeConfig(bytes.NewReader(page))
+		img, err := png.Decode(bytes.NewReader(page))
 		if err != nil {
 			t.Fatalf("rendered page was not a valid PNG: %v", err)
 		}
-		if config.Width == 0 || config.Height == 0 {
-			t.Fatalf("rendered page is %dx%d", config.Width, config.Height)
+		bounds := img.Bounds()
+		if bounds.Dx() == 0 || bounds.Dy() == 0 {
+			t.Fatalf("rendered page is %dx%d", bounds.Dx(), bounds.Dy())
+		}
+
+		// A page missing its fonts renders blank, so require some ink. This is
+		// what catches the proxy image shipping without the PDF base fonts.
+		ink := 0
+		for y := bounds.Min.Y; y < bounds.Max.Y; y++ {
+			for x := bounds.Min.X; x < bounds.Max.X; x++ {
+				r, g, b, _ := img.At(x, y).RGBA()
+				if r < 0x8000 && g < 0x8000 && b < 0x8000 {
+					ink++
+				}
+			}
+		}
+		if ink == 0 {
+			t.Fatal("rendered page has no dark pixels; the proxy image is likely missing PDF fonts (font-liberation)")
 		}
 	})