feat: optional sqlite-vec embedding pipeline for Paid-tier RAG

Dockerfile

@@ -17,6 +17,20 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
 COPY requirements.txt .
 RUN pip install --no-cache-dir -r requirements.txt
 
+# sqlite-vec: optional vector search extension for context embedding (Paid tier)
+RUN set -eux; \
+    SVEC_VER=0.1.6; \
+    ARCH=$(uname -m); \
+    case "$ARCH" in \
+      x86_64)  SVEC_ARCH="x86_64-linux-gnu" ;; \
+      aarch64) SVEC_ARCH="aarch64-linux-gnu" ;; \
+      *)        echo "sqlite-vec: unsupported arch $ARCH — skipping" && exit 0 ;; \
+    esac; \
+    wget -q -O /tmp/sqlite_vec.tar.gz \
+      "https://github.com/asg017/sqlite-vec/releases/download/v${SVEC_VER}/sqlite-vec-${SVEC_VER}-loadable-linux-${SVEC_ARCH}.tar.gz"; \
+    tar -xz -C /usr/lib/python3/ -f /tmp/sqlite_vec.tar.gz --wildcards '*.so' || true; \
+    rm /tmp/sqlite_vec.tar.gz
+
 COPY app/ ./app/
 COPY patterns/ ./patterns/
 COPY scripts/ ./scripts/

app/context/embedder.py

@@ -0,0 +1,64 @@
+"""Ollama embedding client with sqlite-vec storage — BSL licensed."""
+from __future__ import annotations
+
+import logging
+import sqlite3
+import struct
+from pathlib import Path
+
+import httpx
+
+logger = logging.getLogger(__name__)
+
+EMBEDDING_AVAILABLE: bool = False
+
+try:
+    import sqlite_vec  # type: ignore[import]  # noqa: F401
+    EMBEDDING_AVAILABLE = True
+    logger.debug("sqlite-vec loaded — embedding pipeline enabled")
+except ImportError:
+    logger.debug("sqlite-vec not available — embedding pipeline disabled")
+
+
+def embed_chunks(
+    db_path: Path,
+    document_id: str,
+    llm_url: str,
+    model: str = "nomic-embed-text",
+    timeout: float = 60.0,
+) -> int:
+    """Embed all unembedded chunks for a document. Returns count embedded. No-op when EMBEDDING_AVAILABLE is False."""
+    if not EMBEDDING_AVAILABLE:
+        return 0
+
+    conn = sqlite3.connect(str(db_path))
+    conn.execute("PRAGMA journal_mode=WAL")
+    conn.row_factory = sqlite3.Row
+    rows = conn.execute(
+        "SELECT id, text FROM context_chunks WHERE document_id=? AND embedding IS NULL",
+        (document_id,),
+    ).fetchall()
+
+    count = 0
+    for row in rows:
+        try:
+            resp = httpx.post(
+                f"{llm_url.rstrip('/')}/api/embeddings",
+                json={"model": model, "prompt": row["text"]},
+                timeout=timeout,
+            )
+            resp.raise_for_status()
+            vector: list[float] = resp.json().get("embedding") or []
+            if vector:
+                blob = struct.pack(f"{len(vector)}f", *vector)
+                conn.execute(
+                    "UPDATE context_chunks SET embedding=? WHERE id=?",
+                    (blob, row["id"]),
+                )
+                count += 1
+        except Exception as exc:
+            logger.warning("Embedding chunk %s failed: %s", row["id"], exc)
+
+    conn.commit()
+    conn.close()
+    return count

tests/context/test_embedder.py

@@ -0,0 +1,53 @@
+"""Tests for app/context/embedder.py — graceful no-op without sqlite-vec."""
+import sqlite3
+from pathlib import Path
+from unittest.mock import patch
+import pytest
+from app.context import embedder as emb_mod
+
+
+@pytest.fixture
+def db(tmp_path):
+    db_path = tmp_path / "t.db"
+    conn = sqlite3.connect(str(db_path))
+    conn.executescript("""
+        CREATE TABLE context_documents (
+            id TEXT PRIMARY KEY, filename TEXT NOT NULL, doc_type TEXT NOT NULL,
+            full_text TEXT NOT NULL, file_size INTEGER, uploaded_at TEXT NOT NULL
+        );
+        CREATE TABLE context_chunks (
+            id TEXT PRIMARY KEY, document_id TEXT NOT NULL
+                REFERENCES context_documents(id) ON DELETE CASCADE,
+            chunk_index INTEGER NOT NULL, text TEXT NOT NULL, embedding BLOB
+        );
+        INSERT INTO context_documents VALUES ('d1','test.md','markdown','hello',5,'2026-01-01T00:00:00+00:00');
+        INSERT INTO context_chunks VALUES ('c1','d1',0,'hello world',NULL);
+    """)
+    conn.commit()
+    conn.close()
+    return db_path
+
+
+def test_embed_skipped_when_extension_absent(db):
+    with patch.object(emb_mod, "EMBEDDING_AVAILABLE", False):
+        count = emb_mod.embed_chunks(db, "d1", "http://localhost:11434")
+    assert count == 0
+
+
+def test_embed_calls_ollama_when_available(db):
+    import httpx
+
+    class FakeResponse:
+        status_code = 200
+        def raise_for_status(self): pass
+        def json(self): return {"embedding": [0.1, 0.2, 0.3]}
+
+    with patch.object(emb_mod, "EMBEDDING_AVAILABLE", True), \
+         patch("app.context.embedder.httpx.post", return_value=FakeResponse()):
+        count = emb_mod.embed_chunks(db, "d1", "http://localhost:11434")
+    assert count == 1
+    # Verify blob was written
+    conn = sqlite3.connect(str(db))
+    row = conn.execute("SELECT embedding FROM context_chunks WHERE id='c1'").fetchone()
+    conn.close()
+    assert row[0] is not None