Total: 42 USD
`) + got, err := extractFileText(html, "text/html", "invoice.html") + if err != nil { + t.Fatalf("extractFileText() error = %v", err) + } + if !strings.Contains(got, "Invoice") || !strings.Contains(got, "42 USD") { + t.Errorf("extracted text missing expected content: %q", got) + } +} + +// TestExtractFileText_Type3PDF verifies the ToUnicode-based decoding for a PDF +// whose text is drawn with a Type3 font (which tabula's markdown path decodes to +// raw character codes). The fixture maps code 0x01->'H' and 0x02->'i' via a +// ToUnicode CMap and draws <0102>, so correct decoding yields "Hi". +func TestExtractFileText_Type3PDF(t *testing.T) { + got, err := extractFileText(type3PDFFixture(), "application/pdf", "type3.pdf") + if err != nil { + t.Fatalf("extractFileText() error = %v", err) + } + if !strings.Contains(got, "Hi") { + t.Errorf("Type3 PDF did not decode via ToUnicode; got %q", got) + } +} + +func TestInlineFileToText(t *testing.T) { + tests := []struct { + name string + blob *genai.Blob + contains string + }{ + {name: "nil blob", blob: nil, contains: ""}, + { + name: "text file labeled with name", + blob: &genai.Blob{Data: []byte("line1\nline2"), MIMEType: "text/plain", DisplayName: "notes.txt"}, + contains: `Contents of uploaded file "notes.txt"`, + }, + { + name: "unsupported binary returns note", + blob: &genai.Blob{Data: []byte{0x00, 0x01}, MIMEType: "application/zip", DisplayName: "a.zip"}, + contains: "could not be read as text", + }, + } + for _, tt := range tests { + t.Run(tt.name, func(t *testing.T) { + got := InlineFileToText(tt.blob) + if tt.contains == "" { + if got != "" { + t.Errorf("InlineFileToText() = %q, want empty", got) + } + return + } + if !strings.Contains(got, tt.contains) { + t.Errorf("InlineFileToText() = %q, want containing %q", got, tt.contains) + } + }) + } +} diff --git a/go/adk/pkg/fileextract/fixture_test.go b/go/adk/pkg/fileextract/fixture_test.go new file mode 100644 index 0000000000..05aa34d8b8 --- /dev/null +++ b/go/adk/pkg/fileextract/fixture_test.go @@ -0,0 +1,71 @@ +package fileextract + +import ( + "bytes" + "fmt" +) + +// type3PDFFixture builds a minimal, valid PDF whose single page draws text with +// a Type3 font. The font has no embedded glyphs that map to Unicode by encoding; +// the only correct source of text is its ToUnicode CMap (code 0x01->'H', +// 0x02->'i'). The content stream shows <0102>, so a ToUnicode-aware extractor +// yields "Hi" while a naive one yields the raw code bytes. +func type3PDFFixture() []byte { + objects := []string{ + // 1: Catalog + "<< /Type /Catalog /Pages 2 0 R >>", + // 2: Pages + "<< /Type /Pages /Kids [3 0 R] /Count 1 >>", + // 3: Page + "<< /Type /Page /Parent 2 0 R /MediaBox [0 0 300 200] " + + "/Resources << /Font << /F1 4 0 R >> >> /Contents 5 0 R >>", + // 4: Type3 font + "<< /Type /Font /Subtype /Type3 /FontBBox [0 0 750 750] " + + "/FontMatrix [0.001 0 0 0.001 0 0] /FirstChar 1 /LastChar 2 " + + "/Widths [500 500] /CharProcs << /a1 6 0 R /a2 7 0 R >> " + + "/Encoding << /Type /Encoding /Differences [1 /a1 /a2] >> " + + "/ToUnicode 8 0 R >>", + // 5: page content stream + streamObject("BT\n/F1 12 Tf\n10 100 Td\n<0102> Tj\nET\n"), + // 6: CharProc for code 1 + streamObject("500 0 0 0 500 500 d1\n"), + // 7: CharProc for code 2 + streamObject("500 0 0 0 500 500 d1\n"), + // 8: ToUnicode CMap + streamObject(toUnicodeCMap()), + } + + var buf bytes.Buffer + buf.WriteString("%PDF-1.4\n%\xe2\xe3\xcf\xd3\n") + + offsets := make([]int, len(objects)+1) + for i, body := range objects { + offsets[i+1] = buf.Len() + fmt.Fprintf(&buf, "%d 0 obj\n%s\nendobj\n", i+1, body) + } + + xrefStart := buf.Len() + fmt.Fprintf(&buf, "xref\n0 %d\n", len(objects)+1) + buf.WriteString("0000000000 65535 f \n") + for i := 1; i <= len(objects); i++ { + fmt.Fprintf(&buf, "%010d 00000 n \n", offsets[i]) + } + fmt.Fprintf(&buf, "trailer\n<< /Size %d /Root 1 0 R >>\nstartxref\n%d\n%%%%EOF\n", + len(objects)+1, xrefStart) + + return buf.Bytes() +} + +func streamObject(content string) string { + return fmt.Sprintf("<< /Length %d >>\nstream\n%s\nendstream", len(content), content) +} + +func toUnicodeCMap() string { + return "/CIDInit /ProcSet findresource begin\n" + + "12 dict begin\nbegincmap\n" + + "/CMapName /Adobe-Identity-UCS def\n" + + "/CMapType 2 def\n" + + "1 begincodespacerange\n<00>Total: 42 USD
" + out = extract_file_text(html, "text/html", "invoice.html") + assert "Invoice" in out + assert "42 USD" in out diff --git a/python/packages/kagent-adk/tests/unittests/models/test_openai.py b/python/packages/kagent-adk/tests/unittests/models/test_openai.py index 5dba0e52ac..f37670e214 100644 --- a/python/packages/kagent-adk/tests/unittests/models/test_openai.py +++ b/python/packages/kagent-adk/tests/unittests/models/test_openai.py @@ -995,3 +995,56 @@ def test_round_trip_preserves_thought_signature_for_follow_up_tool_result(self): tool_messages = [m for m in messages if m["role"] == "tool"] assert len(tool_messages) == 1 assert tool_messages[0]["extra_content"] == {"google": {"thought_signature": "YWJj"}} + + +class TestConvertContentNonImageFiles: + """Non-image uploads must reach the model as extracted text, not be dropped.""" + + def test_text_file_inline_data_becomes_user_text(self): + contents = [ + Content( + role="user", + parts=[ + Part(text="explain attached invoice"), + Part( + inline_data=types.Blob( + data=b"vendor,amount\nAcme,42", + mime_type="text/csv", + display_name="invoice.csv", + ) + ), + ], + ) + ] + + messages = _convert_content_to_openai_messages(contents) + + user_messages = [m for m in messages if m["role"] == "user"] + assert len(user_messages) == 1 + content = user_messages[0]["content"] + assert isinstance(content, str) + assert "explain attached invoice" in content + assert "invoice.csv" in content + assert "Acme,42" in content + + def test_unsupported_binary_inline_data_adds_note(self): + contents = [ + Content( + role="user", + parts=[ + Part( + inline_data=types.Blob( + data=b"\x00\x01\x02", + mime_type="application/zip", + display_name="archive.zip", + ) + ), + ], + ) + ] + + messages = _convert_content_to_openai_messages(contents) + + user_messages = [m for m in messages if m["role"] == "user"] + assert len(user_messages) == 1 + assert "could not be read as text" in user_messages[0]["content"] diff --git a/ui/src/components/chat/ChatInterface.tsx b/ui/src/components/chat/ChatInterface.tsx index 98b828a54d..80fb274b72 100644 --- a/ui/src/components/chat/ChatInterface.tsx +++ b/ui/src/components/chat/ChatInterface.tsx @@ -2,7 +2,7 @@ import type React from "react"; import { useState, useRef, useEffect, useMemo } from "react"; -import { ArrowBigUp, X, Loader2, Mic, Square } from "lucide-react"; +import { ArrowBigUp, X, Loader2, Mic, Square, Paperclip } from "lucide-react"; import { Button } from "@/components/ui/button"; import { Tooltip, @@ -32,7 +32,8 @@ import { formatA2AClientError } from "@/lib/a2aErrors"; import { useChatRunInSandbox, useChatSubstrateSandbox } from "@/components/chat/ChatAgentContext"; import { v4 as uuidv4 } from "uuid"; import { getStatusPlaceholder, mapA2AStateToStatus } from "@/lib/statusUtils"; -import { Message, DataPart, Task, TaskState } from "@a2a-js/sdk"; +import { Message, DataPart, FilePart, Task, TaskState } from "@a2a-js/sdk"; +import { FILE_ACCEPT, MAX_FILE_BYTES, fileToFilePart, isAllowedFile } from "@/lib/fileUpload"; // Task states where the agent is actively processing — resubscribe to live stream. const RESUBSCRIBE_TASK_STATES: TaskState[] = ["submitted", "working"]; @@ -52,6 +53,8 @@ export default function ChatInterface({ selectedAgentName, selectedNamespace, se const substrateSandbox = useChatSubstrateSandbox(); const router = useRouter(); const containerRef = useRef