pgsemantic 0.2.0__tar.gz → 0.3.0__tar.gz

{pgsemantic-0.2.0 → pgsemantic-0.3.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: pgsemantic
-Version: 0.2.0
+Version: 0.3.0
 Summary: Zero-config semantic search bootstrap for any PostgreSQL database
 Author: Sai Ram Varma Budharaju
 License-Expression: MIT
@@ -49,6 +49,58 @@ It works on your existing tables, keeps embeddings in sync automatically as rows
 
 ---
 
+## See it in action
+
+We loaded 5,980 real clinical trials from ClinicalTrials.gov and searched them in plain English. Here's what the full flow looks like.
+
+**Browse your existing tables — no setup required**
+
+<img src="docs/screenshots/01-tables.png" width="100%" alt="Table Browser showing clinical_trials and news_articles tables with row counts"/>
+
+*pgsemantic connects to your existing database and shows every table. Nothing is modified until you say so.*
+
+---
+
+**Inspect scores your text columns for semantic search suitability**
+
+<img src="docs/screenshots/02-inspect.png" width="100%" alt="Inspect page scoring clinical_trials columns — eligibility and brief_summary rated 3 stars"/>
+
+*Columns are ranked by average text length and content quality. `brief_summary` and `eligibility` score highest — these are the best candidates for embedding.*
+
+---
+
+**Apply sets up everything with one click**
+
+<img src="docs/screenshots/03-apply.png" width="100%" alt="Apply page with column picker checkboxes and embedding model selection"/>
+
+*Pick your columns, choose an embedding model (local runs free on your machine, no API key), click Apply Setup. pgsemantic adds a vector column, HNSW index, and a change-tracking trigger.*
+
+---
+
+**Dashboard shows your embedding coverage**
+
+<img src="docs/screenshots/04-dashboard.png" width="100%" alt="Dashboard showing 5,980 embeddings, 100% coverage, 0 pending"/>
+
+*5,980 rows. 100% coverage. 0 pending. Every clinical trial is now searchable by meaning.*
+
+---
+
+**Search in plain English — results that share zero words with your query**
+
+<img src="docs/screenshots/05-search.png" width="100%" alt="Search results for 'kidney complications in diabetics' returning trials about diabetic nephropathy and DKD"/>
+
+*Query: `kidney complications in diabetics`*
+
+*Top results: trials about **diabetic nephropathy**, **DKD**, **renal insufficiency** — none of which contain the words "kidney complications in diabetics". That's semantic search.*
+
+Results export to CSV or JSON directly from the UI, or from the CLI:
+
+```bash
+pgsemantic search "cardiovascular risk factors" --table clinical_trials --format csv > results.csv
+```
+
+---
+
 ## What is semantic search?
 
 Traditional SQL search finds rows where a column contains the exact words you typed. Semantic search understands *meaning* — it finds results that are conceptually similar even when they share no words with your query.

{pgsemantic-0.2.0 → pgsemantic-0.3.0}/README.md

@@ -12,6 +12,58 @@ It works on your existing tables, keeps embeddings in sync automatically as rows
 
 ---
 
+## See it in action
+
+We loaded 5,980 real clinical trials from ClinicalTrials.gov and searched them in plain English. Here's what the full flow looks like.
+
+**Browse your existing tables — no setup required**
+
+<img src="docs/screenshots/01-tables.png" width="100%" alt="Table Browser showing clinical_trials and news_articles tables with row counts"/>
+
+*pgsemantic connects to your existing database and shows every table. Nothing is modified until you say so.*
+
+---
+
+**Inspect scores your text columns for semantic search suitability**
+
+<img src="docs/screenshots/02-inspect.png" width="100%" alt="Inspect page scoring clinical_trials columns — eligibility and brief_summary rated 3 stars"/>
+
+*Columns are ranked by average text length and content quality. `brief_summary` and `eligibility` score highest — these are the best candidates for embedding.*
+
+---
+
+**Apply sets up everything with one click**
+
+<img src="docs/screenshots/03-apply.png" width="100%" alt="Apply page with column picker checkboxes and embedding model selection"/>
+
+*Pick your columns, choose an embedding model (local runs free on your machine, no API key), click Apply Setup. pgsemantic adds a vector column, HNSW index, and a change-tracking trigger.*
+
+---
+
+**Dashboard shows your embedding coverage**
+
+<img src="docs/screenshots/04-dashboard.png" width="100%" alt="Dashboard showing 5,980 embeddings, 100% coverage, 0 pending"/>
+
+*5,980 rows. 100% coverage. 0 pending. Every clinical trial is now searchable by meaning.*
+
+---
+
+**Search in plain English — results that share zero words with your query**
+
+<img src="docs/screenshots/05-search.png" width="100%" alt="Search results for 'kidney complications in diabetics' returning trials about diabetic nephropathy and DKD"/>
+
+*Query: `kidney complications in diabetics`*
+
+*Top results: trials about **diabetic nephropathy**, **DKD**, **renal insufficiency** — none of which contain the words "kidney complications in diabetics". That's semantic search.*
+
+Results export to CSV or JSON directly from the UI, or from the CLI:
+
+```bash
+pgsemantic search "cardiovascular risk factors" --table clinical_trials --format csv > results.csv
+```
+
+---
+
 ## What is semantic search?
 
 Traditional SQL search finds rows where a column contains the exact words you typed. Semantic search understands *meaning* — it finds results that are conceptually similar even when they share no words with your query.

{pgsemantic-0.2.0 → pgsemantic-0.3.0}/pgsemantic/__init__.py

@@ -1,3 +1,3 @@
 """pgsemantic: Zero-config semantic search bootstrap for PostgreSQL."""
 
-__version__ = "0.2.0"
+__version__ = "0.3.0"

{pgsemantic-0.2.0 → pgsemantic-0.3.0}/pgsemantic/commands/search.py

@@ -1,9 +1,13 @@
 """pgsemantic search — semantic search from the command line."""
 from __future__ import annotations
 
+import csv
+import json
+import sys
+from typing import Any
+
 import typer
 from rich.console import Console
-from rich.panel import Panel
 
 from pgsemantic.config import load_project_config, load_settings
 from pgsemantic.db.client import get_connection
@@ -13,7 +17,28 @@ from pgsemantic.embeddings import get_provider
 console = Console()
 
 # Columns to hide from search results (internal/large)
-_HIDDEN_COLUMNS = frozenset({"embedding", "similarity", "content"})
+_TABLE_HIDDEN_COLUMNS = frozenset({"embedding", "similarity", "content"})
+
+
+def _results_to_csv(results: list[dict[str, Any]]) -> None:
+    """Write results as CSV to stdout."""
+    if not results:
+        return
+    writer = csv.DictWriter(sys.stdout, fieldnames=list(results[0].keys()))
+    writer.writeheader()
+    for row in results:
+        writer.writerow({k: str(v) if v is not None else "" for k, v in row.items()})
+
+
+def _results_to_json(results: list[dict[str, Any]]) -> None:
+    """Write results as a JSON array to stdout."""
+    print(json.dumps(results, indent=2, default=str))
+
+
+def _results_to_jsonl(results: list[dict[str, Any]]) -> None:
+    """Write results as newline-delimited JSON to stdout."""
+    for row in results:
+        print(json.dumps(row, default=str))
 
 
 def search_command(
@@ -39,8 +64,18 @@ def search_command(
         "-d",
         help="Database URL (overrides DATABASE_URL env var).",
     ),
+    fmt: str = typer.Option(
+        "table",
+        "--format",
+        "-f",
+        help="Output format: table (default), csv, json, jsonl.",
+    ),
 ) -> None:
     """Search your database using natural language."""
+    if fmt not in ("table", "csv", "json", "jsonl"):
+        console.print("[red]--format must be one of: table, csv, json, jsonl[/red]")
+        raise typer.Exit(code=1)
+
     settings = load_settings()
     database_url = db if db else settings.database_url
 
@@ -88,9 +123,25 @@ def search_command(
             )
 
         if not results:
-            console.print("\n[yellow]No results found.[/yellow]\n")
-            raise typer.Exit(code=0)
-
+            if fmt == "table":
+                console.print("\n[yellow]No results found.[/yellow]\n")
+            elif fmt == "json":
+                print("[]")
+            # csv and jsonl: empty output is correct
+            raise typer.Exit(code=0)  # re-raised by outer except typer.Exit guard
+
+        # --- machine-readable output ---
+        if fmt == "csv":
+            _results_to_csv(results)
+            return
+        if fmt == "json":
+            _results_to_json(results)
+            return
+        if fmt == "jsonl":
+            _results_to_jsonl(results)
+            return
+
+        # --- default: Rich table output ---
         console.print()
         console.print(
             f"[dim]Results for:[/dim] [cyan]\"{query}\"[/cyan] "
@@ -101,15 +152,13 @@ def search_command(
             sim = float(str(row["similarity"]))
             score_color = "green" if sim >= 0.5 else "yellow" if sim >= 0.3 else "dim"
 
-            # Build metadata line from non-hidden, non-content columns
             meta_parts: list[str] = []
             for key, val in row.items():
-                if key in _HIDDEN_COLUMNS or key == column:
+                if key in _TABLE_HIDDEN_COLUMNS or key == column:
                     continue
                 meta_parts.append(f"[dim]{key}:[/dim] {val}")
             meta_line = "  ".join(meta_parts)
 
-            # Truncate content for preview
             content = str(row.get(column, ""))
             if len(content) > 300:
                 content = content[:300] + "..."

{pgsemantic-0.2.0 → pgsemantic-0.3.0}/pgsemantic/db/vectors.py

@@ -625,6 +625,9 @@ def search_similar(
         quoted_col = _quoted(column)
         sql = SQL_SEARCH_SIMILAR.format(table=qualified, column=quoted_col)
 
+    # ef_search must be >= limit or HNSW returns fewer results than requested.
+    ef_search = max(limit * 2, 100)
+    conn.execute(f"SET hnsw.ef_search = {ef_search}")
     result = conn.execute(
         sql, {"query_vector": query_vector, "limit": limit}
     ).fetchall()

{pgsemantic-0.2.0 → pgsemantic-0.3.0}/pgsemantic/web/app.py

@@ -719,7 +719,7 @@ async def bulk_apply(req: BulkApplyRequest):
 
 
 @app.post("/api/index")
-async def index_table(req: IndexRequest):
+def index_table(req: IndexRequest):
     """Bulk embed existing rows."""
     db_url = _get_db_url()
     settings = load_settings()
@@ -737,15 +737,10 @@ async def index_table(req: IndexRequest):
         provider = get_provider(table_config.model, api_key=api_key)
 
         with get_connection(db_url) as conn:
-            if table_config.storage_mode == "external" and table_config.shadow_table:
-                shadow_qualified = f'"{table_config.schema}"."{table_config.shadow_table}"' if table_config.schema else f'"{table_config.shadow_table}"'
-                row = conn.execute(f"SELECT COUNT(*) AS cnt FROM {shadow_qualified}").fetchone()
-                total = int(str(row["cnt"])) if row else 0
-            else:
-                total = count_total_with_content(
-                    conn, req.table, table_config.column, table_config.schema,
-                    source_columns=table_config.source_columns,
-                )
+            total = count_total_with_content(
+                conn, req.table, table_config.column, table_config.schema,
+                source_columns=table_config.source_columns,
+            )
             already = count_embedded(
                 conn, req.table, table_config.schema,
                 storage_mode=table_config.storage_mode,
@@ -814,7 +809,7 @@ async def index_table(req: IndexRequest):
 
 
 @app.post("/api/reindex")
-async def reindex_table(req: IndexRequest):
+def reindex_table(req: IndexRequest):
     """Re-embed a table by nulling existing embeddings then re-indexing."""
     db_url = _get_db_url()
 
@@ -837,7 +832,7 @@ async def reindex_table(req: IndexRequest):
             conn.commit()
 
         # Now run normal index
-        return await index_table(req)
+        return index_table(req)
 
     except HTTPException:
         raise
@@ -1001,12 +996,7 @@ async def status_dashboard():
             for tc in config.tables:
                 try:
                     embedded = count_embedded(conn, tc.table, schema=tc.schema, storage_mode=tc.storage_mode, shadow_table=tc.shadow_table)
-                    if tc.storage_mode == "external" and tc.shadow_table:
-                        shadow_qualified = f'"{tc.schema}"."{tc.shadow_table}"' if tc.schema else f'"{tc.shadow_table}"'
-                        row = conn.execute(f"SELECT COUNT(*) AS cnt FROM {shadow_qualified}").fetchone()
-                        total = int(str(row["cnt"])) if row else 0
-                    else:
-                        total = count_total_with_content(conn, tc.table, tc.column, schema=tc.schema, source_columns=tc.source_columns)
+                    total = count_total_with_content(conn, tc.table, tc.column, schema=tc.schema, source_columns=tc.source_columns)
                     pending = count_pending(conn, tc.table)
                     failed = count_failed(conn, tc.table)
                     pct = min(round(embedded / total * 100, 1), 100.0) if total > 0 else 0.0

{pgsemantic-0.2.0 → pgsemantic-0.3.0}/pgsemantic/web/static/index.html

@@ -373,7 +373,7 @@
                     <button type="submit" class="btn btn-primary" id="btn-search">Search</button>
                 </form>
             </div></div>
-            <div id="s-results" class="hidden"><div style="display:flex;justify-content:space-between;align-items:center;margin-bottom:12px"><span id="s-title" style="font-weight:600;color:var(--fg-0);font-size:14px"></span><span id="s-count" class="pill pill-blue"></span></div><div id="s-output"></div></div>
+            <div id="s-results" class="hidden"><div style="display:flex;justify-content:space-between;align-items:center;margin-bottom:12px"><span id="s-title" style="font-weight:600;color:var(--fg-0);font-size:14px"></span><div style="display:flex;align-items:center;gap:8px"><span id="s-count" class="pill pill-blue"></span><div id="s-download" class="hidden" style="display:flex;gap:6px"><button class="btn btn-ghost btn-sm" onclick="downloadSearchCSV()">Download CSV</button><button class="btn btn-ghost btn-sm" onclick="downloadSearchJSON()">Download JSON</button></div></div></div><div id="s-output"></div></div>
         </div>
 
         <!-- INSPECT -->
@@ -583,8 +583,10 @@ function saveSearchHistory(t,q){if(!searchHistory[t])searchHistory[t]=[];searchH
 $('s-query')?.addEventListener?.('input',()=>{clearTimeout(searchTimer);searchTimer=setTimeout(()=>{if($('s-query').value.trim().length>=3&&$('s-table').value)doSearch();},300);});
 $('s-table')?.addEventListener?.('change',loadSearchHistory);
 let _searchResults=[];
-async function doSearch(e){if(e)e.preventDefault();const table=$('s-table').value,query=$('s-query').value.trim(),limit=parseInt($('s-limit').value)||5;if(!table||!query)return;const ctrl=startOp('search');setBtn('btn-search',true,'search');$('s-results').classList.remove('hidden');$('s-output').innerHTML='<div class="loading"><span class="spin"></span>Searching...</div>';try{const d=await api('/api/search',{method:'POST',body:JSON.stringify({query,table,limit}),signal:ctrl.signal});saveSearchHistory(table,query);loadSearchHistory();$('s-title').textContent=`Results for "${query}"`;$('s-count').textContent=`${(d.results||[]).length} results`;if(!(d.results||[]).length){$('s-output').innerHTML='<div class="empty-state"><div class="empty-state-icon">&#128270;</div><h3>No results found</h3><p>Try different keywords or check that the table is indexed.</p></div>';return;}_searchResults=d.results;let h='';for(let i=0;i<d.results.length;i++){const r=d.results[i],sim=r.similarity||0;const sc=sim>=0.5?'score-high':sim>=0.3?'score-mid':'score-low';let meta=[];for(const[k,v]of Object.entries(r)){if(k==='similarity'||k==='content')continue;meta.push(`<span style="color:var(--fg-4)">${esc(k)}:</span> ${esc(String(v))}`);}const barColor=sim>=0.5?'var(--green)':sim>=0.3?'var(--amber)':'var(--fg-4)';const barWidth=Math.round(Math.min(sim,1)*100);
-h+=`<div class="result" onclick="openDrawer(buildRowDrawer(_searchResults[${i}]))"><div class="result-head"><span class="result-rank ${sc}">#${i+1}<span class="result-score-bar-wrap"><span class="result-score-bar" style="width:${barWidth}%;background:${barColor}"></span></span>${sim.toFixed(4)}</span><span class="result-meta">${meta.join(' &nbsp; ')}</span></div><div class="result-body">${esc(r.content||'')}</div></div>`;}$('s-output').innerHTML=h;}catch(e){if(e.name==='AbortError')$('s-output').innerHTML='<p style="color:var(--fg-4)">Search cancelled.</p>';else{$('s-output').innerHTML=`<p style="color:var(--red)">${esc(e.message)}</p>`;toast(e.message,false);}}finally{finishOp('search');setBtn('btn-search',false,'search');}}
+function downloadSearchCSV(){if(!_searchResults.length)return;const keys=Object.keys(_searchResults[0]);const rows=[keys.map(k=>'"'+k.replace(/"/g,'""')+'"').join(','),..._searchResults.map(r=>keys.map(k=>{const v=r[k]===null||r[k]===undefined?'':String(r[k]);return'"'+v.replace(/"/g,'""')+'"';}).join(','))];const blob=new Blob([rows.join('\n')],{type:'text/csv'});const a=document.createElement('a');a.href=URL.createObjectURL(blob);a.download=`pgsemantic_search_${Date.now()}.csv`;a.click();setTimeout(()=>URL.revokeObjectURL(a.href),100);}
+function downloadSearchJSON(){if(!_searchResults.length)return;const blob=new Blob([JSON.stringify(_searchResults,null,2)],{type:'application/json'});const a=document.createElement('a');a.href=URL.createObjectURL(blob);a.download=`pgsemantic_search_${Date.now()}.json`;a.click();setTimeout(()=>URL.revokeObjectURL(a.href),100);}
+async function doSearch(e){if(e)e.preventDefault();const table=$('s-table').value,query=$('s-query').value.trim(),limit=parseInt($('s-limit').value)||5;if(!table||!query)return;const ctrl=startOp('search');setBtn('btn-search',true,'search');$('s-results').classList.remove('hidden');$('s-download').classList.add('hidden');$('s-output').innerHTML='<div class="loading"><span class="spin"></span>Searching...</div>';try{const d=await api('/api/search',{method:'POST',body:JSON.stringify({query,table,limit}),signal:ctrl.signal});saveSearchHistory(table,query);loadSearchHistory();$('s-title').textContent=`Results for "${query}"`;$('s-count').textContent=`${(d.results||[]).length} results`;if(!(d.results||[]).length){$('s-download').classList.add('hidden');$('s-output').innerHTML='<div class="empty-state"><div class="empty-state-icon">&#128270;</div><h3>No results found</h3><p>Try different keywords or check that the table is indexed.</p></div>';return;}_searchResults=d.results;$('s-download').classList.remove('hidden');let h='';for(let i=0;i<d.results.length;i++){const r=d.results[i],sim=r.similarity||0;const sc=sim>=0.5?'score-high':sim>=0.3?'score-mid':'score-low';let meta=[];for(const[k,v]of Object.entries(r)){if(k==='similarity'||k==='content')continue;meta.push(`<span style="color:var(--fg-4)">${esc(k)}:</span> ${esc(String(v))}`);}const barColor=sim>=0.5?'var(--green)':sim>=0.3?'var(--amber)':'var(--fg-4)';const barWidth=Math.round(Math.min(sim,1)*100);
+h+=`<div class="result" onclick="openDrawer(buildRowDrawer(_searchResults[${i}]))"><div class="result-head"><span class="result-rank ${sc}">#${i+1}<span class="result-score-bar-wrap"><span class="result-score-bar" style="width:${barWidth}%;background:${barColor}"></span></span>${sim.toFixed(4)}</span><span class="result-meta">${meta.join(' &nbsp; ')}</span></div><div class="result-body">${esc(r.content||'')}</div></div>`;}$('s-output').innerHTML=h;}catch(e){if(e.name==='AbortError'){$('s-download').classList.add('hidden');$('s-output').innerHTML='<p style="color:var(--fg-4)">Search cancelled.</p>';}else{$('s-download').classList.add('hidden');$('s-output').innerHTML=`<p style="color:var(--red)">${esc(e.message)}</p>`;toast(e.message,false);}}finally{finishOp('search');setBtn('btn-search',false,'search');}}
 function buildRowDrawer(row){let h='<div style="display:flex;justify-content:space-between;align-items:center;margin-bottom:16px"><h3>Row Details</h3><button class="btn btn-ghost btn-sm" onclick="closeDrawer()">&#10005;</button></div>';for(const[k,v]of Object.entries(row)){if(k==='similarity'){h+=`<div class="kv-row"><div class="kv-key">Similarity</div><div class="kv-val" style="font-weight:600;font-family:var(--mono)">${Number(v).toFixed(4)}</div></div>`;continue;}h+=`<div class="kv-row"><div class="kv-key">${esc(k)}</div><div class="kv-val">${esc(String(v||''))}</div></div>`;}return h;}
 
 // Inspect
@@ -604,7 +606,7 @@ $('a-table')?.addEventListener?.('change',function(){const val=this.value;const
 async function loadApplyTables(){const sel=$('a-table');if(!sel)return;try{const d=await api('/api/database/tables');const tables=d.tables||[];const cur=sel.value;sel.innerHTML='<option value="">Select a table...</option>';for(const t of tables){const o=document.createElement('option');o.value=t.name;o.textContent=t.name;o.dataset.schema=t.schema||'public';sel.appendChild(o);}if(cur){sel.value=cur;}else{sel.value='';}}catch(e){sel.innerHTML='<option value="">Could not load tables</option>';}}
 let _pendingApplyBody=null;
 async function doApply(e){e.preventDefault();const table=$('a-table').value.trim(),col=$('a-col').value.trim(),cols=$('a-cols').value.trim();const model=$('a-model').value,ext=$('a-ext').checked,schema=$('a-schema').value.trim()||'public';if(!col&&!cols){toast('Specify a column',false);return;}if(col&&cols){toast('Use column OR columns, not both',false);return;}_pendingApplyBody=JSON.stringify({table,column:col,columns:cols,model,external:ext,schema_name:schema});openModal(`<h3>Confirm Apply</h3><div class="dialog-text">This will modify your database:<ul style="margin:8px 0 0 16px"><li>Add an embedding column (or shadow table) to <strong>${esc(table)}</strong></li><li>Create an HNSW index for similarity search</li><li>Install a change-tracking trigger</li></ul></div><div class="dialog-actions"><button class="btn btn-secondary" onclick="closeModal()">Cancel</button><button class="btn btn-primary" onclick="closeModal();execApply()">Confirm</button></div>`);}
-async function execApply(){const bodyStr=_pendingApplyBody;const ctrl=startOp('apply');setBtn('btn-apply',true,'apply');$('a-results').classList.remove('hidden');$('a-results').innerHTML='<div class="card"><div class="card-body"><div class="loading"><span class="spin"></span>Setting up...</div></div></div>';try{const d=await api('/api/apply',{method:'POST',body:bodyStr,signal:ctrl.signal});let h='<div class="card"><div class="card-head"><h3>Setup Complete</h3></div><div class="card-body">';if(d.sql_preview)h+=`<div class="code-block" style="margin-bottom:16px">${esc(d.sql_preview)}</div>`;for(const s of(d.steps||[])){const ok=s.status==='ok'||s.status==='created';const cls=ok?'step-ok':s.status==='already_exists'?'step-skip':'step-fail';const icon=ok?'&#10003;':s.status==='already_exists'?'—':'&#10007;';h+=`<div class="step-row"><span class="step-icon ${cls}">${icon}</span><span>${esc(s.step)}</span><span class="pill ${ok?'pill-green':'pill-amber'}">${esc(s.status)}${s.version?' v'+esc(s.version):''}</span></div>`;}h+=`</div><div class="card-foot"><button class="btn btn-primary btn-sm" onclick="go('index')">Index this table</button><button class="btn btn-secondary btn-sm" onclick="go('dashboard')">Dashboard</button></div></div>`;$('a-results').innerHTML=h;toast(`Semantic search ready on ${d.table}!`);checkConn();}catch(e){if(e.name==='AbortError')$('a-results').innerHTML='<div class="card"><div class="card-body"><p style="color:var(--fg-4)">Setup cancelled.</p></div></div>';else{$('a-results').innerHTML=`<div class="card"><div class="card-body"><p style="color:var(--red)">${esc(e.message)}</p></div></div>`;toast(e.message,false);}}finally{finishOp('apply');setBtn('btn-apply',false,'apply');}}
+async function execApply(){const bodyStr=_pendingApplyBody;const ctrl=startOp('apply');setBtn('btn-apply',true,'apply');$('a-results').classList.remove('hidden');$('a-results').innerHTML='<div class="card"><div class="card-body"><div class="loading"><span class="spin"></span>Setting up...</div></div></div>';try{const d=await api('/api/apply',{method:'POST',body:bodyStr,signal:ctrl.signal});let h='<div class="card"><div class="card-head"><h3>Setup Complete</h3></div><div class="card-body">';if(d.sql_preview)h+=`<div class="code-block" style="margin-bottom:16px">${esc(d.sql_preview)}</div>`;for(const s of(d.steps||[])){const ok=s.status==='ok'||s.status==='created';const cls=ok?'step-ok':s.status==='already_exists'?'step-skip':'step-fail';const icon=ok?'&#10003;':s.status==='already_exists'?'—':'&#10007;';h+=`<div class="step-row"><span class="step-icon ${cls}">${icon}</span><span>${esc(s.step)}</span><span class="pill ${ok?'pill-green':'pill-amber'}">${esc(s.status)}${s.version?' v'+esc(s.version):''}</span></div>`;}h+=`</div><div class="card-foot"><button class="btn btn-primary btn-sm" onclick="go('index');$('x-table').value='${esc(d.table)}'">Index this table →</button><button class="btn btn-secondary btn-sm" onclick="go('dashboard')">Dashboard</button></div></div>`;$('a-results').innerHTML=h;toast('Setup complete — index your rows to enable search');checkConn();}catch(e){if(e.name==='AbortError')$('a-results').innerHTML='<div class="card"><div class="card-body"><p style="color:var(--fg-4)">Setup cancelled.</p></div></div>';else{$('a-results').innerHTML=`<div class="card"><div class="card-body"><p style="color:var(--red)">${esc(e.message)}</p></div></div>`;toast(e.message,false);}}finally{finishOp('apply');setBtn('btn-apply',false,'apply');}}
 
 // Index
 let _indexPollTimer=null;

{pgsemantic-0.2.0 → pgsemantic-0.3.0}/pgsemantic.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: pgsemantic
-Version: 0.2.0
+Version: 0.3.0
 Summary: Zero-config semantic search bootstrap for any PostgreSQL database
 Author: Sai Ram Varma Budharaju
 License-Expression: MIT
@@ -49,6 +49,58 @@ It works on your existing tables, keeps embeddings in sync automatically as rows
 
 ---
 
+## See it in action
+
+We loaded 5,980 real clinical trials from ClinicalTrials.gov and searched them in plain English. Here's what the full flow looks like.
+
+**Browse your existing tables — no setup required**
+
+<img src="docs/screenshots/01-tables.png" width="100%" alt="Table Browser showing clinical_trials and news_articles tables with row counts"/>
+
+*pgsemantic connects to your existing database and shows every table. Nothing is modified until you say so.*
+
+---
+
+**Inspect scores your text columns for semantic search suitability**
+
+<img src="docs/screenshots/02-inspect.png" width="100%" alt="Inspect page scoring clinical_trials columns — eligibility and brief_summary rated 3 stars"/>
+
+*Columns are ranked by average text length and content quality. `brief_summary` and `eligibility` score highest — these are the best candidates for embedding.*
+
+---
+
+**Apply sets up everything with one click**
+
+<img src="docs/screenshots/03-apply.png" width="100%" alt="Apply page with column picker checkboxes and embedding model selection"/>
+
+*Pick your columns, choose an embedding model (local runs free on your machine, no API key), click Apply Setup. pgsemantic adds a vector column, HNSW index, and a change-tracking trigger.*
+
+---
+
+**Dashboard shows your embedding coverage**
+
+<img src="docs/screenshots/04-dashboard.png" width="100%" alt="Dashboard showing 5,980 embeddings, 100% coverage, 0 pending"/>
+
+*5,980 rows. 100% coverage. 0 pending. Every clinical trial is now searchable by meaning.*
+
+---
+
+**Search in plain English — results that share zero words with your query**
+
+<img src="docs/screenshots/05-search.png" width="100%" alt="Search results for 'kidney complications in diabetics' returning trials about diabetic nephropathy and DKD"/>
+
+*Query: `kidney complications in diabetics`*
+
+*Top results: trials about **diabetic nephropathy**, **DKD**, **renal insufficiency** — none of which contain the words "kidney complications in diabetics". That's semantic search.*
+
+Results export to CSV or JSON directly from the UI, or from the CLI:
+
+```bash
+pgsemantic search "cardiovascular risk factors" --table clinical_trials --format csv > results.csv
+```
+
+---
+
 ## What is semantic search?
 
 Traditional SQL search finds rows where a column contains the exact words you typed. Semantic search understands *meaning* — it finds results that are conceptually similar even when they share no words with your query.

{pgsemantic-0.2.0 → pgsemantic-0.3.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "pgsemantic"
-version = "0.2.0"
+version = "0.3.0"
 description = "Zero-config semantic search bootstrap for any PostgreSQL database"
 readme = "README.md"
 license = "MIT"