feat: workspaces, agentic action tools, local Whisper STT, vec recall

- Projects/workspaces: documents grouped into projects; chat RAG scopes to the
  active project. Switcher in the Documents page.
- Agentic action tools: web_search, fetch_url, and remember (first write tool),
  allowlist-gated per playbook.
- Local Whisper STT (faster-whisper, no torch): on-device dictation replacing
  the browser Web Speech API. POST /stt + GET /stt/status; browser fallback.
- Vector index extended to conversation recall (message_vectors), with the
  brute-force cosine scan kept as the fallback.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
jon
2026-07-23 14:22:46 -05:00
co-authored by Claude Opus 4.8
parent f4aea78b55
commit ba6a4ac4e4
12 changed files with 586 additions and 53 deletions
+56
View File
@@ -57,6 +57,62 @@ def test_search_empty_query_returns_nothing():
assert asyncio.run(s.search_documents("", _fake_embed)) == []
def test_conversation_recall_uses_vec_and_matches_brute_force():
s = _store()
if not s.vec_enabled:
import pytest
pytest.skip("sqlite-vec not loadable on this host")
async def run():
s.create_conversation("c1")
s.add_message("c1", "user", "tell me about lego star wars")
s.add_message("c1", "assistant", "lego star wars is a fun game")
s.create_conversation("c2")
s.add_message("c2", "user", "gpu vega vram notes")
s.add_message("c2", "assistant", "vega has 4gb")
hits = await s.semantic_search_conversations("lego star wars", _fake_embed, limit=2, min_score=0.1)
assert hits and hits[0]["id"] == "c1"
conn = s._connect()
n = conn.execute("SELECT COUNT(*) FROM vec_messages").fetchone()[0]
conn.close()
assert n >= 2 # dual-write populated the message vec index
s.vec_enabled = False
bf = await s.semantic_search_conversations("lego star wars", _fake_embed, limit=2, min_score=0.1)
assert bf[0]["id"] == hits[0]["id"]
asyncio.run(run())
def test_projects_scope_documents_and_survive_delete():
s = _store()
async def run():
p = s.create_project("Star Wars")
assert [x["name"] for x in s.list_projects()] == ["Star Wars"]
await s.add_document("Lego", "lego star wars boss tips", _fake_embed, project_id=p["id"])
await s.add_document("GPU", "gpu vega notes", _fake_embed) # unscoped
# project sees only its own; unscoped/all sees both
assert [d["title"] for d in s.list_documents(p["id"])] == ["Lego"]
assert len(s.list_documents(None)) == 2
# scoped search only returns the project's docs
scoped = await s.search_documents("star wars", _fake_embed, min_score=0.1, project_id=p["id"])
assert scoped and all(h["title"] == "Lego" for h in scoped)
# deleting the project keeps the docs but unscopes them
assert s.delete_project(p["id"]) is True
assert s.list_projects() == []
assert len(s.list_documents(None)) == 2
assert len(s.list_documents(p["id"])) == 0 # nothing left in that project
asyncio.run(run())
def test_chunker_overlap_and_hard_split():
s = _store()
# a single oversized paragraph (no blank lines, as in PDF text) is split
+7
View File
@@ -43,6 +43,13 @@ def test_keep_alive_pins_the_model():
assert "keep_alive" not in mgr._apply_keep_alive({"model": "x"})
def test_stt_status_endpoint():
# Reports whether local Whisper is installed; wiring must respond either way.
resp = TestClient(app).get("/stt/status")
assert resp.status_code == 200
assert isinstance(resp.json()["available"], bool)
def test_num_ctx_option_only_when_positive():
# 0 / None -> omit num_ctx so Ollama uses the model default; positive -> set it.
from synapse.ollama_manager import _chat_options
+18
View File
@@ -23,6 +23,24 @@ async def _drain(gen):
return [s async for s in gen]
def test_remember_writes_a_memory_fact(tmp_path, monkeypatch):
# The `remember` action tool persists a fact through the store.
import synapse.memory.store as store_mod
from synapse.memory.store import PersistentMemoryStore
fresh = PersistentMemoryStore(tmp_path / "m.db")
monkeypatch.setattr(tools, "store", fresh)
out = asyncio.run(tools.dispatch("remember", {"text": "user likes tea", "section": "Prefs"}))
assert "user likes tea" in out
assert any(it.text == "user likes tea" for it in fresh.all())
def test_action_tools_registered():
for name in ("web_search", "fetch_url", "remember"):
assert name in tools.REGISTRY
names = [s["function"]["name"] for s in tools.schemas_for(["web_search", "remember", "nope"])]
assert names == ["web_search", "remember"]
class _FakeManager:
"""Returns a tool_call on the first chat() call, plain content after."""
def __init__(self):