HugBrowse v1.0 — Full Test Specification
Component
Details
OS Windows 11
HugBrowse v1.0.0 (Tauri + React, dev mode)
Ollama Running on :11434, models: gemma3:1b (815MB Q4_K_M), qwen3-embedding:8b (4.7GB Q4_K_M)
LM Studio Installed (not running — start if needed)
API Server Auto-starts on :8080
MCP Server hugbrowse-mcp-server v1.0.0 (12 tools, stdio)
GPU Check via Resource Monitor
TS-01: App Launch & Onboarding
ID
Test
Steps
Expected
01.1
Cold start Run npm run tauri:dev
App window opens, no crash, no panic
01.2
API auto-start Check Developer tab after launch
API server shows "Running" on port 8080
01.3
Onboarding flow If first run: complete 4-step wizard
Hardware detected → Browse → Download → Chat
01.4
Privacy consent Should show privacy modal if not accepted
Modal dismissible, persists choice
01.5
System tray Minimize app
Tray icon appears, right-click shows menu (Open, Quick Chat, Quit)
TS-02: Model Search & Discovery
ID
Test
Steps
Expected
02.1
Basic search Type "llama 3" in search bar, press Enter
Results load with model cards showing name, author, downloads, likes
02.2
Filter by task Click "text-generation" filter
Results filtered to text-generation models only
02.3
Sort options Change sort to "Most Likes"
Results reorder by like count descending
02.4
Infinite scroll Scroll to bottom of results
More results load automatically
02.5
Empty search Search for "xyznonexistent12345"
"No results found" message shown
02.6
Click model card Click any model result
Navigates to ModelDetailPage with full info
ID
Test
Steps
Expected
03.1
Model info loads Navigate to a model (e.g. search "bartowski/Llama-3.2-3B-Instruct-GGUF")
Shows model name, author, tags, download count, likes
03.2
README renders Scroll to README section
Model card markdown renders correctly
03.3
Files table Click "Files" tab
Shows file names (NOT blank), sizes, types
03.4
GGUF files highlighted Look at GGUF files in the list
GGUF files are identifiable/highlighted
03.5
Download Now section Look for "Download Now" section
Prominent section with bold title, border, download buttons
03.6
Compatibility check Look for hardware compatibility info
Shows if model fits in your RAM/VRAM
ID
Test
Steps
Expected
04.1
Start download Click download on a small GGUF file (e.g. ~800MB gemma-like model)
Download starts, progress bar appears
04.2
Progress tracking Watch download progress
Shows %, speed (MB/s), ETA
04.3
Pause download Click pause button
Download pauses, .part file exists on disk
04.4
Resume download Click resume button
Download resumes from where it left off
04.5
Cancel download Start a new download, then cancel
.part file cleaned up, status shows "Cancelled"
04.6
Completed download Let a download finish
Status shows "Complete", SHA-256 validated
04.7
Persistence Close and reopen app
Completed downloads still shown in history
TS-05: Model Loading (Local Sidecar)
ID
Test
Steps
Expected
05.1
Load from imports Go to Chat → select a downloaded GGUF → Load
Status: unloaded → loading → running
05.2
GPU detection Check GPU layers setting before loading
Shows available GPUs (NVIDIA/AMD/Intel)
05.3
Context size config Set context size to 2048 before loading
Model loads with specified context
05.4
Check API ready After load completes
/v1/models returns the loaded model
05.5
Unload model Click unload button
Status returns to "unloaded", llama-server process killed
05.6
Memory warning Try to load a model that exceeds available VRAM
Warning shown with memory estimates
TS-06: Backend Management
ID
Test
Steps
Expected
06.1
List backends Go to Settings or Chat backend selector
Shows configured backends
06.2
Add Ollama backend Add custom URL backend: http://127.0.0.1:11434/v1, name "Ollama"
Backend added, shows in list
06.3
Test Ollama connection Click "Test Connection" on Ollama backend
Shows "Online", latency_ms, discovers model name
06.4
Set active backend Select Ollama as active backend
Active indicator updates
06.5
Existing Ollama entry Check backends.json loaded the "Ollama: qwen3-embedding:8b" entry
Should show with custom_url type, URL http://127.0.0.1:11434/v1
06.6
Remove backend Delete a test backend
Removed from list
TS-07: Chat (Core Feature — CRITICAL)
ID
Test
Steps
Expected
07.1
Empty state Open Chat with no model loaded
Shows "Get Started — Load a Model" with 3-step guide
07.2
Chat with Ollama Set Ollama backend active → type "Hello, who are you?" → Send
Model responds with text, streaming visible
07.3
Chat with local sidecar Load a GGUF → set local backend → send message
Response streams in real-time
07.4
Multi-turn Send follow-up message in same session
Context maintained, coherent response
07.5
New session Click "New Chat"
Fresh session, no prior messages
07.6
Session list Check sidebar
Shows all chat sessions with titles
07.7
System prompt Set system prompt to "You are a pirate" → send message
Response matches the persona
07.8
Token speed Check message metadata after response
Shows tokens/sec metric
07.9
Edit message Click edit on a user message → modify → resend
Conversation rewinds and regenerates
07.10
Delete message Delete an assistant message
Message removed from session
07.11
Backend attribution Check each message
Shows which backend generated the response
07.12
Long response Ask "Write a 500 word essay about AI"
Full response streams without truncation or crash
ID
Test
Steps
Expected
08.1
Page loads Navigate to Resource Monitor
Shows CPU, RAM, GPU, disk metrics
08.2
Real-time updates Watch for 10 seconds
Values update periodically
08.3
Hardware tier Check tier classification
Shows tier (potato/laptop/gaming/workstation/server)
08.4
Under load Load a model → monitor resources
RAM/VRAM usage increases visibly
TS-09: Settings & Configuration
ID
Test
Steps
Expected
09.1
Theme toggle Switch between light/dark/system theme
Theme changes immediately
09.2
HF token Enter a Hugging Face token → save
Token persisted in secure store
09.3
Presets Create an inference preset (GPU layers=20, ctx=4096)
Preset saved, selectable in chat
09.4
Export config Click export settings
JSON file downloaded
09.5
Import config Import previously exported config
Settings restored
09.6
MCP servers View configured MCP servers in settings
Shows registered servers
TS-10: Developer Page & API Server
ID
Test
Steps
Expected
10.1
API status Open Developer tab
Shows "Running" on port 8080 (auto-started)
10.2
Status endpoint curl http://127.0.0.1:8080/api/v1/statusReturns {"status":"ok","version":"1.0.0",...}
10.3
Models endpoint curl http://127.0.0.1:8080/v1/modelsReturns OpenAI-format model list
10.4
Chat completions curl -X POST http://127.0.0.1:8080/v1/chat/completions -d '{"model":"...","messages":[...]}'Returns chat response (requires model loaded)
10.5
Stop/restart server Click Stop → Start
Server stops and restarts cleanly
10.6
Request log Make a few API calls → check log
Shows request history with timestamps
10.7
Auth tokens Create an API token
Token generated, shown once
10.8
Auth enforcement Enable auth → call API without token
Returns 401 Unauthorized
TS-11: MCP Server (AI Agent Integration)
ID
Test
Steps
Expected
11.1
Server starts MCP server visible in VS Code Copilot Chat
Shows as connected/available
11.2
health_check Call hugbrowse_health_check
Returns: API online, HF accessible, models count
11.3
search_models Call hugbrowse_search_models(query="phi 3")
Returns model list with metadata
11.4
find_gguf_models Call hugbrowse_find_gguf_models(query="llama 3 8b")
Returns GGUF repos with file sizes
11.5
get_model_detail Call hugbrowse_get_model_detail(model_id="microsoft/Phi-3-mini-4k-instruct-gguf")
Returns full metadata
11.6
list_model_files Call hugbrowse_list_model_files(model_id="...", gguf_only=true)
Returns only GGUF files
11.7
get_readme Call hugbrowse_get_readme(model_id="...")
Returns model card markdown
11.8
server_status Call hugbrowse_server_status
Returns running/port/config
11.9
list_loaded_models Call hugbrowse_list_loaded_models
Returns loaded instances (or empty)
11.10
load_model Call hugbrowse_load_model(model_path="...", model_name="test")
Model loads, returns success with port
11.11
chat Call hugbrowse_chat(message="What is 2+2?")
Returns model response
11.12
multi_turn_chat Call with messages array including prior context
Returns contextual response
11.13
unload_model Call hugbrowse_unload_model(instance_id="...")
Model unloaded, resources freed
TS-12: Model Import & Scan
ID
Test
Steps
Expected
12.1
Scan Ollama Trigger Ollama model scan
Finds gemma3:1b and qwen3-embedding:8b
12.2
Scan LM Studio Trigger LM Studio scan (if models exist)
Finds any cached GGUF files
12.3
Manual import Select a GGUF file from disk
File validated and added to imports
12.4
Import persistence Close and reopen app
Imported models still listed
TS-13: Recommended Models Page
ID
Test
Steps
Expected
13.1
Page loads Navigate to Recommended
Shows model suggestions based on hardware tier
13.2
Task categories Check for Chat, Code, Image categories
Models organized by task type
13.3
Hardware-appropriate Verify suggested models
Sizes match your hardware capabilities
TS-14: Deep Links & Navigation
ID
Test
Steps
Expected
14.1
All routes work Navigate to each page via sidebar
All 12 pages load without errors
14.2
Back/forward Use browser back/forward buttons
Navigation history works
14.3
Keyboard shortcuts Test any registered shortcuts
Shortcuts trigger correct actions
14.4
Quick Chat from tray Right-click tray → Quick Chat
App opens to Chat page
TS-15: Error Handling & Edge Cases
ID
Test
Steps
Expected
15.1
Offline mode Disconnect internet → search models
Graceful error, "You're offline" indicator
15.2
Backend down Stop Ollama → try to chat via Ollama backend
Clear error message, not a crash
15.3
Invalid model path Try to load non-existent GGUF path
Error message with path shown
15.4
Port conflict Try to start API server on occupied port
Error message about port in use
15.5
Disk full Simulate low disk space during download
Warning shown before download starts
15.6
Rapid actions Click load/unload rapidly
No race conditions, app stays stable
ID
Test
Steps
Expected
16.1
App startup time Measure cold start to window visible
< 5 seconds
16.2
Search latency Measure time from search to results
< 3 seconds
16.3
Model load time Measure load time for gemma3:1b (~800MB)
< 30 seconds
16.4
First token latency Send message → time to first token
< 2 seconds (local), < 5 seconds (remote)
16.5
Token throughput Check tokens/sec during generation
Reported accurately in UI
16.6
Memory stability Run chat for 10 minutes
No memory leaks (RAM stays stable)
Quick-Run Test Script (MCP Tools) Run these in order in Copilot Chat to test the MCP integration end-to-end:
1. hugbrowse_health_check()
2. hugbrowse_search_models(query="gemma", filter="gguf", limit=3)
3. hugbrowse_find_gguf_models(query="phi 3 mini", limit=2)
4. hugbrowse_get_model_detail(model_id="microsoft/Phi-3-mini-4k-instruct-gguf")
5. hugbrowse_get_readme(model_id="microsoft/Phi-3-mini-4k-instruct-gguf")
6. hugbrowse_list_model_files(model_id="microsoft/Phi-3-mini-4k-instruct-gguf", gguf_only=true)
7. hugbrowse_server_status()
8. hugbrowse_list_loaded_models()
If you have a downloaded GGUF file, continue:
9. hugbrowse_load_model(model_path="<path-to-gguf>", model_name="Test Model")
10. hugbrowse_list_loaded_models()
11. hugbrowse_chat(message="Hello! What is 2+2?")
12. hugbrowse_multi_turn_chat(messages=[{role:"user",content:"What is the capital of France?"},{role:"assistant",content:"Paris"},{role:"user",content:"What about Germany?"}])
13. hugbrowse_unload_model(instance_id="<from step 10>")
Priority
Sections
Rationale
P0 — Must Pass TS-01, TS-07, TS-11
App must launch, chat must work, MCP must function
P1 — Critical TS-02, TS-03, TS-04, TS-05, TS-06, TS-10
Core user flows: search, download, load, backends, API
P2 — Important TS-08, TS-09, TS-12, TS-15
Monitor, settings, imports, error handling
P3 — Nice to have TS-13, TS-14, TS-16
Recommendations, navigation, performance
Total: 16 sections, 85 test cases
Generated: 2026-03-09
Target: HugBrowse v1.0.0