Compare commits

..

36 Commits

Author SHA1 Message Date
tyler ad8f1397bc Fixed intent in the dev server 2026-05-19 15:20:06 -04:00
Tyler bda9e5a462 Fixing Intent in Dev.py 2026-05-19 14:30:42 -04:00
TySS-Dev f471140bd6 Upload files to "dev" 2026-05-19 06:01:19 -04:00
TySS-Dev 68ff90b655 Upload files to "/" 2026-05-19 06:01:01 -04:00
TySS-Dev 6daf947d00 Delete directory 'tests' 2026-05-19 03:09:34 -04:00
tyler af3e74c92a Moved dev.py to dev/ 2026-05-19 03:09:18 -04:00
tyler 7e06e07e4f moved an import line to the top 2026-05-19 03:08:57 -04:00
tyler 8b7c8f7df8 Adding base64 import back 2026-05-19 02:59:01 -04:00
tyler 93d263cdc3 Cleaning up code 2026-05-19 02:56:01 -04:00
tyler b914d13d4e Removing the boarder around model dropdown 2026-05-19 02:48:10 -04:00
tyler eee0fd8709 Revert ollama_answers.py to theme-aware state before border animation 2026-05-19 02:46:56 -04:00
tyler baec4522cf Fixing a animation around the input box 2026-05-19 02:43:27 -04:00
tyler 1702d9cd20 Fixing a animation around the input box 2026-05-19 02:40:42 -04:00
tyler 2a5a501a96 Fixing a animation around the input box 2026-05-19 02:39:00 -04:00
tyler 64aa62f5e0 Fixing a animation around the input box 2026-05-19 02:35:05 -04:00
tyler 378a485ba7 Adding a animation around the input box 2026-05-19 02:32:53 -04:00
tyler f66264b92a Attempting to make elements theme aware 2026-05-19 02:23:26 -04:00
tyler ff3b75d129 Attempting to make elements theme aware 2026-05-19 02:20:49 -04:00
tyler 08d4915d4a Attempting to make elements theme aware 2026-05-19 02:16:12 -04:00
tyler ce42f9a652 Attempting to make elements theme aware 2026-05-19 02:12:18 -04:00
tyler 9e784c8b8b Attempting to make elements theme aware 2026-05-19 02:07:25 -04:00
tyler 8e7752c2de Updated Issues in README 2026-05-19 02:07:01 -04:00
tyler 78941479db Reworking css 2026-05-19 01:49:18 -04:00
tyler 83494bb023 Reworking our injection 2026-05-19 00:05:50 -04:00
tyler e46c752aec Maybe working divider 2026-05-19 00:02:29 -04:00
tyler 541d98f7f1 Maybe working divider 2026-05-18 23:56:23 -04:00
tyler 4c749b825c Fixing conversation history and couldn't figure out how to remove SearXNG info box so just adding a smart divider 2026-05-18 23:53:04 -04:00
tyler 23ecac6afa Fixed mayber 2026-05-18 23:14:22 -04:00
Tyler 4b36a261c4 Attempting to fix conversation history 2026-05-18 15:11:18 -04:00
TySS-Dev eeac7fcd88 Added more known issues 2026-05-17 20:13:19 -04:00
TySS-Dev 1c3824b7a4 Fixed typo 2026-05-17 20:01:43 -04:00
TySS-Dev a7c031d27b Fixed check boxes 2026-05-17 20:00:56 -04:00
TySS-Dev 5e2b2a246f Added known issues and roadmap 2026-05-17 19:59:55 -04:00
TySS-Dev ffad0de8ae Fixed flow diagram 2026-05-17 19:51:19 -04:00
TySS-Dev 3dffeb384b Fixed a typo in README 2026-05-17 19:46:04 -04:00
TySS-Dev 85d1481bd9 Updated README 2026-05-17 19:45:37 -04:00
7 changed files with 1798 additions and 488 deletions
+1
View File
@@ -3,6 +3,7 @@ __pycache__/
*$py.class
venv/
.env
dev/.env
.idea/
.vscode/
.agent/
+82 -34
View File
@@ -10,44 +10,50 @@
A SearXNG plugin that generates local AI overviews powered by Ollama, using search results as RAG context.
Features:
- Token-by-token UI streaming
- Clickable inline citations
- Interactive mode: continue summary, ask follow-ups, copy, or regenerate
- Simple response mode with no extras
- Internally called low-latency RAG for follow-ups (bypasses HTTP loopback)
- Native network integration via `searx.network` (respects proxy/SSL settings)
- Stateless conversation persistence/shareability via URL hash
- Model selector in the AI overview widget
- Does not slow down result loading
- One file install
- Real-time streaming via Valkey — responses stream token by token using a background thread + Valkey job queue, working around granian's broken generator support for true streaming feel
- TF-IDF result reranking — fetched page content is scored against the query using BM25-style TF-IDF before being sent to Ollama, surfacing the most relevant sources first
- Smart chunking — pages are split into 512-token overlapping segments and the highest-scoring chunk per page is selected for context
- Intent detection — queries are automatically classified into 8 intent types (factual, howto, technical, comparison, opinion, current, local, general) with tailored system prompts per type
- Conversation memory — 30-minute cross-search conversation history stored in Valkey, so follow-up questions work even after navigating to a new search
- Markdown rendering — AI responses render bold, italic, lists, headers, and inline code natively in the result box
- Intent emoji badge — a small emoji appears next to "AI Overview" indicating the detected query type
## Features:
- Inline numbered citations
- Interactive mode - Continue summary, ask follow-ups, copy, or regenerate
- Overview of ranked results with prompts based on detected query intent:
- `How To` `Technical` `Factual` `Comparison` `Opinion` `Current` `Local` `Geneal`
- Internally called RAG for follow-ups
- Native network integration via `searx.network`
- Stateless conversation presistence/shareability via URL hash
- Ollama model selector
- Feeds fetched results to Ollama without slowing down SearXNG results
- Real-time streaming via Valkey (No waiting for a completed response)
- TF-IDF result ranking before being sent to Ollama
- Smart chunking - Pages are split into 512-token segments and highest-scoring chunk per page used for context
- Conversation memory - 30-minute cross-search conversation history via Valkey for follow-up questions
- Markdown support
- Intent emoji badge showing what intent prompt was used
## Install
1. Download the plugin:
### Main repo (Gitea)
```bash
curl -o ollama_answers.py https://raw.githubusercontent.com/TySP-Dev/ollama-ai-answers-searxng/master/ollama_answers.py
curl -o ollama_answers.py https://git.tysstech.com/TySS-Dev/ollama-ai-answers-searxng/raw/branch/main/ollama_answers.py
```
2. Copy to your SearXNG plugins directory:
### Mirror repo (Github):
```bash
cp ollama_answers.py ~/searxng/plugins/ollama_answers.py
curl -o ollama_answers.py https://raw.githubusercontent.com/TySP-Dev/ollama-ai-answers-searxng/main/ollama_answers.py
```
3. Add the volume mount to your `docker-compose.yml` under the searxng service:
3. Copy to your SearXNG plugins directory:
```bash
cp ollama_answers.py path_to/searxng/plugins/ollama_answers.py
```
4. Add the volume mount to your `docker-compose.yml` under the searxng service:
```yaml
volumes:
- ./plugins/ollama_answers.py:/usr/local/searxng/searx/plugins/ollama_answers.py:Z
```
4. Add environment variables to `docker-compose.yml`:
5. Add environment variables to `docker-compose.yml`:
```yaml
environment:
- LLM_URL=http://ollama:11434/v1/chat/completions
@@ -55,14 +61,14 @@ Features:
- VALKEY_HOST=searxng-valkey
```
5. Add to `settings.yml` plugins section:
6. Add to `settings.yml` plugins section:
```yaml
plugins:
searx.plugins.ollama_answers.SXNGPlugin:
active: true
```
6. Restart SearXNG:
7. Restart SearXNG:
```bash
docker compose up -d --force-recreate core
```
@@ -96,6 +102,48 @@ Configure via environment variables.
6. Client-side script calls a signed endpoint (`/ai-stream`)
7. Ollama streams a response token-by-token in the UI
## Known Issues
- [ ] Update README with all updates
- [x] When asking a follow up question the previous output disappears
- [x] Parts of the UI are not theme aware resulting in a unpolished look when not using a dark theme
- [x] When SearXNG provides a info blob for a search it appears on top of the overview i.e. `Wikipedia` or `Linux`
For any issues not stated here please create an issue ticket on [Gitea](https://git.tysstech.com/TySS-Dev/ollama-ai-answers-searxng/issues) or [GitHub](https://github.com/TySP-Dev/ollama-ai-answers-searxng/issues) and add the `bug` tag.
## Roadmap
### Dev Server
- [x] Stream viewer — show tokens arriving in real time in the debug panel as they come out of Valkey, so you can see exactly what the LLM is generating chunk by chunk
- [x] TF-IDF score visualizer — show a table of which URLs were fetched, their scores, and which chunks were selected for context
- [ ] Intent detection display — show what intent was detected and which system prompt was used for each query
- [ ] Saved queries — save/load test queries so you can quickly re-run the same set of searches after making changes to the plugin
- [ ] A/B model comparison — run the same query against two different models simultaneously and show both responses side by side
- [ ] Response time breakdown — show how long each phase took: SearXNG fetch, page fetching, TF-IDF scoring, LLM stream start, stream complete
- [ ] Context inspector — show the full assembled context string that gets sent to the LLM, so you can see exactly what it's working with
- [ ] Prompt viewer — show the full system prompt + user prompt that gets sent to Ollama
- [ ] Export button — copy the full context + prompt + response as a JSON blob for bug reports
- [ ] Per-intent system prompt editor — edit the system prompts for each intent type live without restarting
- [ ] Token counter — show estimated token count of the context being sent
### Plugin
- [ ] Working on feature plans
## Architecture
```
@@ -105,35 +153,35 @@ Configure via environment variables.
└────────────────┬────────────────────────────────────┘
┌────────────────▼────────────────────────────────────┐
│ SearXNG + Plugin
│ post_search()
│ SearXNG + Plugin │
│ │
│ post_search() │
│ → _enrich_results() ← ThreadPoolExecutor │
│ → _fetch_page_text() × 5 parallel │
│ → _chunk_text() + _tfidf_score() │
│ → rerank by score │
│ → _assemble_context() │
│ → inject AI Overview HTML + JS │
│ /ai-stream
│ → validate token
│ │
│ /ai-stream │
│ → validate token │
│ → _detect_intent() → select system prompt │
│ → _load_conversation() from Valkey │
│ → launch stream_to_valkey() thread │
│ → return {job_id} immediately │
│ │
│ stream_to_valkey() [background thread] │
│ → Ollama stream=True │
│ → RPUSH tokens to Valkey │
│ → RPUSH __DONE__ when complete │
│ │
│ /ai-status/{job_id} │
│ → LRANGE chunks from offset │
│ → return {chunks, done} │
└────────────────┬────────────────────────────────────┘
┌────────────────▼────────────────────────────────────┐
│ Valkey
│ Valkey │
│ ai:job:{id}:chunks (list, TTL 120s) │
│ ai:job:{id}:status (string, TTL 120s) │
│ ai:conv:{session} (JSON, TTL 1800s) │
+39
View File
@@ -0,0 +1,39 @@
# AI Answers Plugin — Dev Server Config
# Copy this to .env and fill in your values
# .env is gitignored and never committed
# Ollama endpoint (required)
LLM_URL=http://localhost:11434/v1/chat/completions
# Default model
LLM_MODEL=qwen3.5:9b
# Max response tokens
LLM_MAX_TOKENS=200
# Response temperature (0.0 - 2.0)
LLM_TEMPERATURE=0.2
# Bearer token for authenticated LLM endpoints
# Leave empty if no Bearer token is needed for your server
LLM_API_KEY=
# Live SearXNG instance for real search results
# Leave empty to use mock results
SEARXNG_URL=
# Valkey for streaming (required)
# Start with: docker run -d --name dev-valkey -p 6379:6379 valkey/valkey:9-alpine
VALKEY_HOST=localhost
VALKEY_PORT=6379
# Dev server host and port
DEV_HOST=127.0.0.1
DEV_PORT=5000
# Plugin settings
LLM_INTERACTIVE=true
LLM_QUESTION_MARK_REQUIRED=false
LLM_TABS=general,science,it,news
LLM_CONTEXT_DEEP_COUNT=5
LLM_CONTEXT_SHALLOW_COUNT=15
+1482
View File
File diff suppressed because it is too large Load Diff
+192 -108
View File
@@ -1,12 +1,11 @@
import json, os, logging, base64, time, hashlib, re, http.client, ssl, concurrent.futures, threading, math
import json, os, logging, base64, typing, time, hashlib, re, http.client, ssl, concurrent.futures, threading, math
from collections import Counter
from urllib.parse import urlparse
from searx import network
try:
from searx.network import get_network
except ImportError:
get_network = None
from flask import Response, request, abort, jsonify
from flask import request, abort, jsonify
from searx.plugins import Plugin, PluginInfo
from searx.result_types import EngineResults
from searx import settings
@@ -24,7 +23,6 @@ except ImportError:
logger.warning("AI Answers: valkey package not found. Streaming via Valkey unavailable.")
TOKEN_EXPIRY_SEC = 3600
STREAM_CHUNK_SIZE = 512
STREAM_TIMEOUT_SEC = 60
CONV_TTL = 1800
@@ -276,17 +274,17 @@ INTERACTIVE_CSS = '''
width: 32px;
height: 32px;
padding: 0;
border: none;
border: 1px solid var(--color-result-border, rgba(0,0,0,0.1));
border-radius: 4px;
background: var(--color-sidebar-bg, #424247);
color: var(--color-search-url, #bbb);
background: var(--color-base-background-hover, rgba(0,0,0,0.06));
color: var(--color-base-font, inherit);
cursor: pointer;
vertical-align: middle;
line-height: 1.4;
}
.sxng-btn:hover {
background: var(--color-search-url, #303033);
color: var(--color-sidebar-bg, #bbb);
background: var(--color-result-border, rgba(0,0,0,0.15));
color: var(--color-base-font, inherit);
}
.sxng-btn svg { width: 18px; height: 18px; fill: currentColor; }
.sxng-input-wrapper {
@@ -300,9 +298,9 @@ INTERACTIVE_CSS = '''
.sxng-input {
width: 100%;
height: -webkit-fill-available;
background: var(--color-sidebar-bg, #424247);
border: none;
color: var(--color-base-font, #cdd6f4);
background: var(--color-base-background-hover, rgba(0,0,0,0.06));
border: 1px solid var(--color-result-border, rgba(0,0,0,0.15));
color: var(--color-base-font, inherit);
font-family: -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, Helvetica, Arial, sans-serif;
font-size: 0.78em;
padding: 3px 8px;
@@ -311,7 +309,7 @@ INTERACTIVE_CSS = '''
vertical-align: middle;
}
.sxng-input:focus { outline: none; }
.sxng-input::placeholder { color: var(--color-base-font, #333); opacity: 0.35; }
.sxng-input::placeholder { color: var(--color-base-font, inherit); opacity: 0.4; }
.sxng-input-line {
position: absolute;
bottom: 0;
@@ -335,23 +333,24 @@ INTERACTIVE_CSS = '''
opacity: 0.55;
animation: sxng-fade-in-up 0.3s ease-out forwards;
}
.sxng-input-wrapper:focus-within {
opacity: 1;
color: var(--color-result-link, #5e81ac);
.sxng-input-wrapper:focus-within {
opacity: 1;
color: var(--color-result-link, #5e81ac);
background: var(--color-base-background-hover, rgba(0,0,0,0.05)) !important;
}
.sxng-model-select {
appearance: none;
-webkit-appearance: none;
background: url("data:image/svg+xml;charset=UTF-8,%3C%3Fxml%20version%3D%221.0%22%20encoding%3D%22UTF-8%22%3F%3E%0A%3Csvg%20xmlns%3D%22http%3A%2F%2Fwww.w3.org%2F2000%2Fsvg%22%20viewBox%3D%220%200%20512%20512%22%3E%0A%3Cg%20fill%3D%22%23aaa%22%3E%0A%3Cpolygon%20points%3D%22128%2C192%20256%2C320%20384%2C192%22%2F%3E%3C%2Fg%3E%0A%3C%2Fsvg%3E") calc(100% + 2rem) / 1rem no-repeat content-box border-box;
background-color: #424247;
background-color: var(--color-base-background-hover, rgba(0,0,0,0.06));
text-overflow: ellipsis;
border-width: 0 2rem 0 0;
border-color: transparent;
border: 0px solid var(--color-result-border, rgba(0,0,0,0.1));
border-right-width: 2rem;
border-right-color: transparent;
border-radius: 5px;
outline: none;
height: 25px;
color: var(--color-search-url, #bbb);
color: var(--color-base-font, inherit);
font-size: .9rem;
padding: 1px 10px 1px 10px !important;
margin: 0;
@@ -361,8 +360,7 @@ INTERACTIVE_CSS = '''
vertical-align: middle;
}
.sxng-model-select:hover {
background-color: #303033;
color: var(--color-search-url, #bbb);
background-color: var(--color-result-border, rgba(0,0,0,0.15));
}
.sxng-reasoning {
margin: 0.5rem 0; padding: 0.5rem;
@@ -385,7 +383,8 @@ INTERACTIVE_CSS = '''
font-size: 0.75em;
color: var(--color-result-link, #5e81ac);
text-decoration: none;
opacity: 0.75;
opacity: 1;
font-weight: 600;
}
.sxng-citation-item a:hover {
opacity: 1;
@@ -395,18 +394,18 @@ INTERACTIVE_CSS = '''
margin-bottom: 0.75rem;
padding: 0.5rem;
border-left: 2px solid var(--color-result-link, #5e81ac);
opacity: 0.6;
opacity: 0.85;
font-size: 0.85em;
}
.sxng-prior-history summary {
cursor: pointer;
color: var(--color-result-link, #5e81ac);
font-weight: 600;
font-weight: 700;
}
.sxng-prior-answer {
margin: 0.25rem 0;
padding-left: 0.5rem;
color: var(--color-base-font, #cdd6f4);
color: var(--color-base-font, inherit);
}
.sxng-md-content {
line-height: 1.6;
@@ -507,7 +506,7 @@ CITATION_HELPER_JS = r'''
const re = /\[(\d{1,2}(?:\s*,\s*\d{1,2})*)\]/g;
let lastIdx = 0;
const matches = [...text.matchAll(re)];
matches.forEach(match => {
if (match.index > lastIdx) {
const s = document.createElement('span');
@@ -542,7 +541,7 @@ CITATION_HELPER_JS = r'''
});
lastIdx = match.index + match[0].length;
});
if (lastIdx < text.length) {
const s = document.createElement('span');
s.className = 'sxng-chunk';
@@ -600,23 +599,6 @@ INTERACTIVE_JS = r'''
_ms.appendChild(_o);
}
}
if (window.getComputedStyle && box) {
try {
const docStyles = getComputedStyle(document.documentElement);
let accent = docStyles.getPropertyValue('--color-result-link').trim();
if (!accent) {
const a = document.createElement('a');
document.body.appendChild(a);
accent = getComputedStyle(a).color;
document.body.removeChild(a);
}
if (accent) {
box.style.setProperty('--color-result-link', accent);
box.style.setProperty('--sxng-ai-accent', accent);
}
} catch(e) {}
}
// conversation saved as base64 URL fragment.
const updateState = () => {
try {
@@ -636,13 +618,13 @@ INTERACTIVE_JS = r'''
}
return btoa(bin);
};
let b64 = encodeB64(state);
while (b64.length > 2000 && state.t.length > 2) {
state.t.splice(1, 2); // Delete in Q&A pairs
b64 = encodeB64(state);
}
history.replaceState(null, null, '#ai=' + b64);
} catch(e) {}
};
@@ -658,17 +640,17 @@ INTERACTIVE_JS = r'''
if (state.u && Array.isArray(state.u)) {
urls = state.u;
}
conversation.turns = state.t.map(t => ({
role: t.r === 'u' ? 'user' : 'assistant',
content: t.c.trim(),
ts: 0
}));
const injectCitations = (text) => {
return renderCitations(text, urls);
};
data.innerHTML = '';
conversation.turns.forEach((turn, i) => {
if (turn.role === 'user') {
@@ -686,7 +668,6 @@ INTERACTIVE_JS = r'''
}
});
box.style.display = 'block';
if(wrapper) wrapper.style.display = '';
if(footer && is_interactive) footer.style.display = 'flex';
restored = true;
}
@@ -705,7 +686,18 @@ INTERACTIVE_JS = r'''
};
document.getElementById('btn-regen').onclick = async () => {
data.innerHTML = '<span class="sxng-cursor"></span>';
// Remove only the last assistant response and its citation footer
const lastMd = [...data.querySelectorAll('.sxng-md-content')].pop();
if (lastMd) {
const nextSib = lastMd.nextElementSibling;
if (nextSib && nextSib.classList.contains('sxng-citation-footer')) nextSib.remove();
lastMd.remove();
}
const existingCursor = data.querySelector('.sxng-cursor');
if (existingCursor) existingCursor.remove();
const regenCursor = document.createElement('span');
regenCursor.className = 'sxng-cursor';
data.appendChild(regenCursor);
footer.style.display = 'none';
if (conversation.turns.length > 0 && conversation.turns[conversation.turns.length - 1].role === 'assistant') {
@@ -745,10 +737,10 @@ INTERACTIVE_JS = r'''
const handleAction = async (e) => {
if (e) e.preventDefault();
const val = input.value.trim();
conversation.turns.push({role: 'user', content: val, ts: Date.now()});
updateState();
const currentText = conversation.turns.slice(0, -1).slice(-6)
.map(t => (t.role === 'user' ? 'Q' : 'A') + ': ' + t.content)
.join('\\n\\n');
@@ -771,7 +763,7 @@ INTERACTIVE_JS = r'''
const newCursor = document.createElement('span');
newCursor.className = 'sxng-cursor';
data.appendChild(newCursor);
const synthesized = synthesizeQuery(q_init, val);
let auxContext = null;
try {
@@ -788,7 +780,7 @@ INTERACTIVE_JS = r'''
}
}
} catch (err) {}
await startStream(val, currentText, auxContext);
updateState();
} else {
@@ -860,16 +852,92 @@ FRONTEND_JS_TEMPLATE = r"""
const conversation = {
originalQuery: q_init,
originalContext: new TextDecoder().decode(Uint8Array.from(atob(b64_init), c => c.charCodeAt(0))),
originalSources: [...urls],
turns: [{role: 'user', content: q_init, ts: Date.now()}]
};
const box = document.getElementById('sxng-stream-box');
const data = document.getElementById('sxng-stream-data');
const wrapper = box.closest('.answer');
if (wrapper) wrapper.style.display = 'none';
(function applyTheme() {
try {
const root = document.documentElement;
const s = getComputedStyle(root);
const get = (v, fallback) => s.getPropertyValue(v).trim() || fallback;
const theme = {
'--color-answer-background': get('--color-answer-background', '#313338'),
'--color-answer-font': get('--color-answer-font', '#fff'),
'--color-result-link': get('--color-result-link', '#8aacf7'),
'--color-base-font': get('--color-base-font', '#cdd6f4'),
'--color-sidebar-bg': get('--color-sidebar-bg', '#424247'),
'--color-result-hover': get('--color-result-hover', '#303033'),
'--color-base-background': get('--color-base-background', '#2a2a2e'),
'--color-search-font': get('--color-search-font', '#bbb'),
'--color-result-border': get('--color-result-border', '#4c566a'),
'--color-result-description':get('--color-result-description', '#d8dee9'),
'--color-toolkit-select-background': get('--color-toolkit-select-background', '#313338'),
};
// Apply to box and any ai-answers container
const targets = [box, document.getElementById('ai-answers')].filter(Boolean);
targets.forEach(el => {
Object.entries(theme).forEach(([k, v]) => {
if (v) el.style.setProperty(k, v);
});
});
} catch(e) {}
})();
// Move AI Overview outside #answers, place it before #results
(function relocateBox() {
const answersDiv = document.getElementById('answers');
if (!box || !answersDiv) return;
// Create our own container
const aiContainer = document.createElement('div');
aiContainer.id = 'ai-answers';
const rootStyle = getComputedStyle(document.documentElement);
const getVar = (v, fb) => rootStyle.getPropertyValue(v).trim() || fb;
const bg = getVar('--color-answer-background', '');
const answerFont = getVar('--color-answer-font', '');
// Detect light mode by checking if answer font is dark
const isLight = answerFont && (answerFont.includes('0,0,0') ||
answerFont.includes('#000') || answerFont.includes('#333') ||
answerFont.includes('#444') || answerFont.includes('rgb(0') ||
answerFont.includes('rgb(3') || answerFont.includes('rgb(4') ||
answerFont.includes('rgb(5') || answerFont.includes('rgb(6'));
const containerBg = isLight
? 'rgba(0,0,0,0.06)'
: (bg || 'var(--color-answer-background, #313338)');
aiContainer.style.cssText = [
`background: ${containerBg}`,
'padding: 1rem',
'margin: 0 0 1rem 0',
`color: ${getVar('--color-answer-font', 'var(--color-answer-font, #fff)')}`,
'border-radius: 8px',
'box-sizing: border-box',
'width: 100%'
].join('; ');
// Move our box into the new container
aiContainer.appendChild(box);
const resultsGrid = document.getElementById('results');
if (resultsGrid) {
// Insert as first child of #results grid so grid-area:answers applies
resultsGrid.insertBefore(aiContainer, resultsGrid.firstChild);
} else {
answersDiv.parentNode.insertBefore(aiContainer, answersDiv);
}
// Hide #answers entirely since our box is now elsewhere
answersDiv.style.display = 'none';
})();
let restored = false;
let isStreaming = false;
__CITATION_HELPER_JS__
(function applyIntentBadge() {
@@ -932,11 +1000,10 @@ FRONTEND_JS_TEMPLATE = r"""
console.warn('[AI Answers] Stream already in progress, ignoring duplicate call');
return;
}
isStreaming = true;
try {
const ctx = auxContext || conversation.originalContext;
if (wrapper) wrapper.style.display = '';
box.style.display = 'block';
const controller = new AbortController();
@@ -991,6 +1058,11 @@ FRONTEND_JS_TEMPLATE = r"""
data.appendChild(cursor);
}
const streamContainer = document.createElement('div');
streamContainer.className = 'sxng-stream-container';
if (cursor) cursor.before(streamContainer);
else data.appendChild(streamContainer);
let buffer = '';
let fullText = '';
const flushBuffer = (force = false) => {
@@ -998,8 +1070,7 @@ FRONTEND_JS_TEMPLATE = r"""
if (force) {
const fragment = renderCitations(buffer, urls);
if (cursor) cursor.before(fragment);
else data.appendChild(fragment);
streamContainer.appendChild(fragment);
buffer = '';
return;
}
@@ -1014,12 +1085,12 @@ FRONTEND_JS_TEMPLATE = r"""
const s = document.createElement('span');
s.className = 'sxng-chunk';
s.textContent = preText;
cursor.before(s);
streamContainer.appendChild(s);
}
const citationText = match[0];
const fragment = renderCitations(citationText, urls);
cursor.before(fragment);
streamContainer.appendChild(fragment);
buffer = buffer.substring(match.index + match[0].length);
}
@@ -1030,7 +1101,7 @@ FRONTEND_JS_TEMPLATE = r"""
const s = document.createElement('span');
s.className = 'sxng-chunk';
s.textContent = buffer;
cursor.before(s);
streamContainer.appendChild(s);
buffer = '';
}
} else {
@@ -1039,7 +1110,7 @@ FRONTEND_JS_TEMPLATE = r"""
const s = document.createElement('span');
s.className = 'sxng-chunk';
s.textContent = safeChunk;
cursor.before(s);
streamContainer.appendChild(s);
}
buffer = buffer.substring(openIdx);
@@ -1047,7 +1118,7 @@ FRONTEND_JS_TEMPLATE = r"""
const s = document.createElement('span');
s.className = 'sxng-chunk';
s.textContent = buffer[0];
cursor.before(s);
streamContainer.appendChild(s);
buffer = buffer.substring(1);
}
}
@@ -1109,15 +1180,9 @@ FRONTEND_JS_TEMPLATE = r"""
}
}
streamContainer.remove();
if (cursor) cursor.remove();
// Replace streamed text nodes with markdown-rendered content
Array.from(data.childNodes).forEach(node => {
if (node.nodeType !== 1 || !node.classList.contains('sxng-prior-history')) {
node.remove();
}
});
const rendered = parseMarkdown(fullText.trim());
const mdDiv = document.createElement('div');
mdDiv.className = 'sxng-md-content';
@@ -1144,13 +1209,13 @@ FRONTEND_JS_TEMPLATE = r"""
console.error('[AI Answers] Fatal stream exception:', e);
const errSpan = document.createElement('span');
errSpan.style.cssText = 'color: #bf616a; font-weight: bold; display: block; margin-top: 0.5rem;';
if (e.name === 'AbortError') {
errSpan.textContent = "⚠️ Connection to AI provider timed out.";
} else {
errSpan.textContent = "⚠️ AI Widget encountered a fatal error. Check browser console.";
}
if (data) {
const cursor = data.querySelector('.sxng-cursor');
if (cursor) cursor.remove();
@@ -1281,8 +1346,6 @@ INTENT_CONFIGS = {
},
}
import typing
if typing.TYPE_CHECKING:
from searx.search import SearchWithPlugins
from searx.extended_types import SXNG_Request
@@ -1372,7 +1435,7 @@ class SXNGPlugin(Plugin):
'content': str(ib.get('content') or '')[:2000],
'attributes': ib.get('attributes', [])
})
answers = []
for a in list(raw_answers)[:2]:
ans_text = ""
@@ -1382,7 +1445,7 @@ class SXNGPlugin(Plugin):
ans_text = str(a['answer'])
if ans_text and 'id="sxng-stream-box"' not in ans_text and not ans_text.strip().startswith('<'):
answers.append(ans_text)
return results, infoboxes, answers
def init(self, app):
@@ -1390,10 +1453,10 @@ class SXNGPlugin(Plugin):
def ai_auxiliary_search():
if not self.api_key:
abort(403)
data = request.json or {}
token = data.get('tk', '')
# Token access control
try:
ts, sig = token.rsplit('.', 1)
@@ -1413,13 +1476,13 @@ class SXNGPlugin(Plugin):
offset = data.get('offset', 0)
if not query:
return jsonify({'results': []})
try:
from searx.search import SearchWithPlugins
from searx.search.models import SearchQuery
from searx.query import RawTextQuery
from searx.webadapter import get_engineref_from_category_list
preferences = getattr(request, 'preferences', None)
disabled_engines = preferences.engines.get_disabled() if preferences else []
rtq = RawTextQuery(query, disabled_engines)
@@ -1427,7 +1490,7 @@ class SXNGPlugin(Plugin):
category_list = [c.strip() for c in categories.split(',') if c.strip()]
else:
category_list = categories or ['general']
enginerefs = get_engineref_from_category_list(category_list, disabled_engines)
sq = SearchQuery(
query=rtq.getQuery(),
@@ -1437,19 +1500,19 @@ class SXNGPlugin(Plugin):
)
search_obj = SearchWithPlugins(sq, request, user_plugins=[])
result_container = search_obj.search()
raw_results = result_container.get_ordered_results()
raw_infoboxes = getattr(result_container, 'infoboxes', [])
raw_answers = getattr(result_container, 'answers', [])
results, infoboxes, answers = self._parse_aux_results(raw_results, raw_infoboxes, raw_answers)
context_str, new_urls = self._assemble_context(results, infoboxes, answers, offset)
return jsonify({
'context': context_str,
'new_urls': new_urls,
'results': results,
'results': results,
'infoboxes': infoboxes,
'answers': answers,
'query': query
@@ -1662,6 +1725,16 @@ class SXNGPlugin(Plugin):
job_id = hashlib.sha256(f"{time.time()}{q}".encode()).hexdigest()[:16]
# Persist intent for dev UI
logger.warning(f"INTENT BEFORE PERSIST: {repr(intent)}")
logger.warning(f"JOB_ID BEFORE PERSIST: {repr(job_id)}")
try:
vk = _get_valkey()
vk.setex(f"ai:job:{job_id}:intent", 3600, intent)
logger.debug(f"{PLUGIN_NAME}: persisted intent '{intent}' for job {job_id}")
except Exception:
logger.exception(f"{PLUGIN_NAME}: failed to persist intent")
payload_dict = {
"model": effective_model,
"messages": [
@@ -1869,12 +1942,12 @@ class SXNGPlugin(Plugin):
"""Builds context string from normalized search data. Returns (context_str, urls)."""
context_parts = []
result_urls = []
knowledge_graph_lines = []
for ib in infoboxes:
ib_name = ib.get('name', '') or ib.get('infobox', '') or ib.get('title', '')
ib_content = str(ib.get('content', '')).replace('\n', ' ').strip()
if ib_name:
parts = [f"INFOBOX [{ib_name}]:"]
if ib_content:
@@ -1884,16 +1957,16 @@ class SXNGPlugin(Plugin):
attr_value = attr.get('value', '')
if attr_label and attr_value:
parts.append(f" {attr_label}: {attr_value}")
knowledge_graph_lines.append(" ".join(parts) if len(parts) == 2 else "\n".join(parts))
for ans_text in answers:
if ans_text and not str(ans_text).startswith('<'):
knowledge_graph_lines.append(f"ANSWER: {str(ans_text)[:300]}")
if knowledge_graph_lines:
context_parts.append("KNOWLEDGE GRAPH:\n" + "\n".join(knowledge_graph_lines))
deep_lines = []
for i, r in enumerate(clean_results[:self.context_deep_count]):
url = r.get('url', '')
@@ -1909,10 +1982,10 @@ class SXNGPlugin(Plugin):
logger.debug(f"{PLUGIN_NAME}: falling back to snippet for [{idx}] {domain}")
content = str(r.get('content', '')).replace('\n', ' ').strip()[:800]
deep_lines.append(f"[{idx}] {domain}{date_str}: {title}: {content}")
if deep_lines:
context_parts.append("DEEP SOURCES:\n" + "\n".join(deep_lines))
if self.context_shallow_count > 0:
shallow_lines = []
start_idx = self.context_deep_count
@@ -1924,10 +1997,10 @@ class SXNGPlugin(Plugin):
title = r.get('title', '').replace('\n', ' ').strip()[:60]
idx = i + 1 + start_idx + offset
shallow_lines.append(f"[{idx}] {domain}: {title}")
if shallow_lines:
context_parts.append("SHALLOW SOURCES (headlines):\n" + "\n".join(shallow_lines))
return "\n\n".join(context_parts), result_urls
def post_search(self, request: "SXNG_Request", search: "SearchWithPlugins") -> EngineResults:
@@ -1951,7 +2024,7 @@ class SXNGPlugin(Plugin):
raw_results = search.result_container.get_ordered_results()
raw_infoboxes = getattr(search.result_container, 'infoboxes', [])
raw_answers = getattr(search.result_container, 'answers', [])
q_clean = search.search_query.query.strip()
clean_results, infoboxes, answers = self._parse_aux_results(raw_results, raw_infoboxes, raw_answers)
clean_results = self._enrich_results(clean_results, q_clean)
@@ -1974,12 +2047,23 @@ class SXNGPlugin(Plugin):
detected_intent = _detect_intent(q_clean)
js_intent = safe_json(detected_intent)
# Persist intent for dev tooling / UI
try:
vk = _get_valkey()
vk.setex(
f"ai:job:{job_id}:intent",
1800,
detected_intent
)
except Exception as e:
logger.debug(f"{PLUGIN_NAME}: failed to persist intent: {e}")
b64_context = base64.b64encode(context_str.encode('utf-8')).decode('utf-8')
total_context_count = self.context_deep_count + self.context_shallow_count
raw_urls = [r.get('url', '') for r in clean_results[:total_context_count]]
js_q = safe_json(q_clean)
js_lang = safe_json(lang)
js_urls = safe_json(raw_urls)
@@ -2021,7 +2105,7 @@ class SXNGPlugin(Plugin):
.replace("__JS_Q__", js_q)
html_payload = f'''
<article id="sxng-stream-box" class="answer" style="display:none; margin-top: 0; margin-bottom: 0;">
<article id="sxng-stream-box" class="answer" style="display:none; margin: 0; padding: 0;">
<style>
@keyframes sxng-fade-pulse {{
0%, 100% {{ opacity: 0.1; }}
@@ -2071,11 +2155,11 @@ class SXNGPlugin(Plugin):
</style>
<div class="sxng-ai-header">
<span class="sxng-ai-label">
<span style="color:#4a9eff;font-size:1.1em;">✦</span> AI Overview
<span style="color:var(--color-result-link, #4a9eff);font-size:1.1em;">✦</span> AI Overview
</span>
<select id="sxng-model-select" class="sxng-model-select" title="Select model"></select>
</div>
<p id="sxng-stream-data" style="white-space: pre-wrap; color: var(--color-result-description); font-size: 0.95rem; margin:0;"><span class="sxng-cursor"></span></p>
<p id="sxng-stream-data" style="white-space: pre-wrap; color: var(--color-answer-font, var(--color-result-description, inherit)); font-size: 0.95rem; margin:0;"><span class="sxng-cursor"></span></p>
{interactive_html}
<script>
{js_code}
@@ -2085,4 +2169,4 @@ class SXNGPlugin(Plugin):
search.result_container.answers.add(results.types.Answer(answer=Markup(html_payload)))
except Exception as e:
logger.error(f"{PLUGIN_NAME}: {e}")
return results
return results
+2
View File
@@ -1,3 +1,5 @@
flask
flask-babel
certifi
python-dotenv
valkey
-346
View File
@@ -1,346 +0,0 @@
import sys
import os
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
import logging
from types import ModuleType
from flask import Flask, request, redirect
logging.basicConfig(level=logging.INFO, format='%(levelname)s: %(message)s')
os.environ.setdefault('LLM_URL', 'http://localhost:11434/v1/chat/completions')
# SearXNG module mocks
searx = ModuleType("searx")
searx_plugins = ModuleType("searx.plugins")
searx_results = ModuleType("searx.result_types")
class MockPlugin:
def __init__(self, cfg):
self.active = getattr(cfg, 'active', True)
class MockPluginInfo:
def __init__(self, **kwargs):
self.meta = kwargs
class MockEngineResults:
def __init__(self):
self.types = ModuleType("types")
self.types.Answer = lambda *args, **kwargs: kwargs.get('answer', args[0] if args else "")
self._results = []
def add(self, res):
self._results.append(res)
searx_plugins.Plugin = MockPlugin
searx_plugins.PluginInfo = MockPluginInfo
searx_results.EngineResults = MockEngineResults
searx.settings = {'server': {'secret_key': 'demo-secret'}}
searx.network = ModuleType("searx.network")
sys.modules["searx"] = searx
sys.modules["searx.plugins"] = searx_plugins
sys.modules["searx.result_types"] = searx_results
# Network module mock
searx_network = ModuleType("searx.network")
def mock_network_call(method, url, **kwargs):
import http.client, ssl, json
from urllib.parse import urlparse
parsed = urlparse(url)
port = parsed.port or (443 if parsed.scheme=='https' else 80)
target = f"{parsed.hostname}:{port}"
if parsed.scheme == 'https':
conn = http.client.HTTPSConnection(target, timeout=30, context=ssl.create_default_context())
else:
conn = http.client.HTTPConnection(target, timeout=30)
headers = kwargs.get('headers', {})
body = None
if kwargs.get('json'):
body = json.dumps(kwargs['json'])
elif kwargs.get('data'):
body = kwargs['data']
path = parsed.path
if parsed.query:
path += f"?{parsed.query}"
if kwargs.get('params'):
from urllib.parse import urlencode
query_str = urlencode(kwargs['params'])
if '?' in path:
path += f"&{query_str}"
else:
path += f"?{query_str}"
conn.request(method, path, body=body, headers=headers)
return conn.getresponse()
def mock_stream(method, url, **kwargs):
res = mock_network_call(method, url, **kwargs)
class MockResponse:
def __init__(self, r):
self.status_code = r.status
self.text = "Mock Response" # Stub
self._r = r
def generator():
while True:
chunk = res.read(128)
if not chunk: break
yield chunk
return MockResponse(res), generator()
def mock_get(url, **kwargs):
import json
res = mock_network_call('GET', url, **kwargs)
class MockResponse:
def __init__(self, r):
self.status_code = r.status
self._content = r.read()
self.text = self._content.decode('utf-8')
def json(self):
return json.loads(self.text)
return MockResponse(res)
searx_network.stream = mock_stream
searx_network.get = mock_get
sys.modules["searx.network"] = searx_network
from ollama_answers import SXNGPlugin
from flask_babel import Babel
app = Flask(__name__)
babel = Babel(app)
class MockConfig:
active = True
plugin = SXNGPlugin(MockConfig())
plugin.init(app)
@app.route("/config", methods=["POST"])
def update_config():
url = request.form.get("url", "").strip()
bearer = request.form.get("bearer", "").strip()
model = request.form.get("model", "").strip()
query = request.form.get("q", "")
if url:
plugin.endpoint_url = url
plugin.api_key = bearer if bearer else "ollama"
if model:
plugin.model = model
redirect_q = f"?q={query}" if query else ""
return redirect(f"/{redirect_q}")
@app.route("/search")
def mock_search():
query = request.args.get("q", "")
format_type = request.args.get("format", "html")
if format_type != "json":
return "Demo only supports JSON format", 400
results = [
{"title": f"Result 1 for: {query}", "content": f"This is simulated content about {query}. It contains relevant information.", "url": f"https://example.com/1/{query.replace(' ', '-')}", "publishedDate": "2026-01-18"},
{"title": f"Result 2 for: {query}", "content": f"Additional information regarding {query}. More context and details.", "url": f"https://example.com/2/{query.replace(' ', '-')}", "publishedDate": "2026-01-17"},
{"title": f"Result 3 for: {query}", "content": f"Further reading on {query}. Expert analysis.", "url": f"https://example.com/3/{query.replace(' ', '-')}", "publishedDate": "2026-01-16"},
]
return {
"results": results,
"infoboxes": [],
"answers": [],
"suggestions": [f"{query} explained", f"{query} tutorial"]
}
@app.route("/")
def index():
query = request.args.get("q", "why is the sky blue")
class MockSearchQuery:
pageno = 1
lang = 'en'
categories = ['general']
MockSearchQuery.query = query
class MockSearch:
search_query = MockSearchQuery()
class MockResultContainer:
def __init__(self):
self.answers = set()
def get_ordered_results(self):
base_results = [
{"title": "Wikipedia", "content": "The sky appears blue due to Rayleigh scattering of sunlight. When sunlight enters the atmosphere, it collides with gas molecules and scatters in all directions. Blue light scatters more than other colors because it travels in shorter waves.", "url": "https://en.wikipedia.org/wiki/Rayleigh_scattering", "publishedDate": "2026-01-15"},
{"title": "NASA Science", "content": "Shorter blue wavelengths scatter more than longer red wavelengths. This phenomenon, discovered by Lord Rayleigh in the 1870s, explains why we see a blue sky during the day.", "url": "https://science.nasa.gov/blue-sky", "publishedDate": "2026-01-10"},
{"title": "Physics Today", "content": "The atmosphere acts as a filter, scattering blue light in all directions while letting other colors pass through more directly.", "url": "https://physicstoday.org/atmosphere", "publishedDate": "2026-01-01"},
{"title": "Scientific American", "content": "At sunset, light travels through more atmosphere, scattering away the blue and leaving reds and oranges.", "url": "https://scientificamerican.com/sunset", "publishedDate": "2025-12-20"},
{"title": "National Geographic", "content": "Ocean color also results from light scattering and absorption by water molecules.", "url": "https://nationalgeographic.com/ocean-blue", "publishedDate": "2025-12-15"},
]
broad_results = [
{"title": "MIT OpenCourseWare: Atmospheric Physics", "content": "Course materials.", "url": "https://ocw.mit.edu/physics"},
{"title": "NOAA: Understanding the Atmosphere", "content": "Educational resource.", "url": "https://noaa.gov/atmosphere"},
{"title": "BBC Science: Why is the sky blue?", "content": "Explainer article.", "url": "https://bbc.com/science/sky"},
{"title": "Khan Academy: Light and Color", "content": "Video lesson.", "url": "https://khanacademy.org/light"},
{"title": "HowStuffWorks: Rayleigh Scattering", "content": "Detailed explanation.", "url": "https://howstuffworks.com/rayleigh"},
{"title": "Physics Stack Exchange: Sky color discussion", "content": "Q&A thread.", "url": "https://physics.stackexchange.com/sky"},
{"title": "Quora: Atmospheric optics explained", "content": "Community answers.", "url": "https://quora.com/atmosphere"},
]
if 'quantum' in query.lower():
return [
{"title": "IBM Quantum", "content": "Quantum computers rely on qubits, which can represent 0, 1, or both via superposition. They solve complex problems faster.", "url": "https://www.ibm.com/quantum", "publishedDate": "2026-01-15"},
{"title": "Nature Physics", "content": "Entanglement allows qubits to be correlated instantly across distances. This is key for quantum cryptography and teleportation.", "url": "https://nature.com/articles/quantum", "publishedDate": "2026-01-10"},
{"title": "Wikipedia", "content": "Quantum computing uses quantum mechanics. Major applications include drug discovery and materials science.", "url": "https://en.wikipedia.org/wiki/Quantum_computing", "publishedDate": "2025-12-01"}
] + broad_results
return base_results + broad_results
result_container = MockResultContainer()
search = MockSearch()
plugin.post_search(None, search)
injection_html = ""
if search.result_container.answers:
injection_html = list(search.result_container.answers)[0]
bearer_display = plugin.api_key if plugin.api_key != "ollama" else ""
return f"""
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<title>AI Answers Demo</title>
<style>
body {{
font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', sans-serif;
padding: 2rem;
max-width: 800px;
margin: 0 auto;
background: #2e3440;
color: #eceff4;
}}
:root {{
--color-result-border: #3b4252;
--color-result-description: #d8dee9;
--color-base-font: #88c0d0;
--color-result-link: #81a1c1;
}}
.meta {{ color: #81a1c1; font-size: 0.9rem; }}
hr {{ border-color: #4c566a; }}
a {{ color: #88c0d0; }}
.config-panel {{
background: #3b4252;
border-radius: 6px;
padding: 1rem 1.25rem;
margin-bottom: 1.25rem;
}}
.config-panel summary {{
cursor: pointer;
font-size: 0.85rem;
color: #81a1c1;
user-select: none;
}}
.config-panel summary:hover {{ color: #88c0d0; }}
.config-row {{
display: flex;
flex-direction: column;
gap: 0.5rem;
margin-top: 0.75rem;
}}
.config-row label {{
font-size: 0.8rem;
color: #81a1c1;
}}
.config-row input {{
background: #2e3440;
border: 1px solid #4c566a;
border-radius: 4px;
color: #eceff4;
font-size: 0.85rem;
padding: 0.4rem 0.6rem;
width: 100%;
box-sizing: border-box;
}}
.config-row input:focus {{ outline: none; border-color: #81a1c1; }}
.config-btn {{
margin-top: 0.75rem;
background: #4c566a;
border: none;
border-radius: 4px;
color: #eceff4;
cursor: pointer;
font-size: 0.85rem;
padding: 0.4rem 1rem;
}}
.config-btn:hover {{ background: #5e81ac; }}
.search-row {{
display: flex;
gap: 0.5rem;
margin-bottom: 1.25rem;
}}
.search-row input {{
flex: 1;
background: #3b4252;
border: 1px solid #4c566a;
border-radius: 4px;
color: #eceff4;
font-size: 0.95rem;
padding: 0.45rem 0.75rem;
}}
.search-row input:focus {{ outline: none; border-color: #81a1c1; }}
.search-row button {{
background: #5e81ac;
border: none;
border-radius: 4px;
color: #eceff4;
cursor: pointer;
font-size: 0.9rem;
padding: 0.45rem 1rem;
}}
.search-row button:hover {{ background: #81a1c1; }}
</style>
</head>
<body>
<details class="config-panel" {'open' if not bearer_display and 'localhost' in plugin.endpoint_url else ''}>
<summary>&#9881; Ollama Configuration</summary>
<form method="POST" action="/config">
<input type="hidden" name="q" value="{query}">
<div class="config-row">
<label>Endpoint URL</label>
<input type="text" name="url" value="{plugin.endpoint_url}" placeholder="http://localhost:11434/v1/chat/completions">
</div>
<div class="config-row">
<label>Bearer Token <span style="opacity:0.5;">(optional)</span></label>
<input type="text" name="bearer" value="{bearer_display}" placeholder="Leave empty if not required">
</div>
<button type="submit" class="config-btn">Apply</button>
</form>
</details>
<form class="search-row" method="GET" action="/">
<input type="text" name="q" value="{query}" placeholder="Ask something...">
<button type="submit">Search</button>
</form>
<p class="meta">Model: <strong>{plugin.model}</strong></p>
<hr>
{injection_html if injection_html else '<p style="color:#f66;">No response — check your Ollama endpoint and token above.</p>'}
</body>
</html>
"""
if __name__ == "__main__":
print("AI Answers - Demo\n")
print(f" Endpoint: {plugin.endpoint_url}")
print(f" Model: {plugin.model or 'N/A'}")
print(f" Mode: {'interactive' if plugin.interactive else 'simple'}")
print(f"\n http://localhost:5000/?q=why+is+the+sky+blue\n")
app.run(debug=False, port=5000)