feat: playbook tools, RAG, vision, voice, chat controls, faster boot

- Tool-using playbooks: read-only tool registry (search_memory,
  search_history, search_documents, list_models, get_time), per-playbook
  allowlist, agentic loop, and a "running tool" status indicator.
- Document ingest / RAG: documents table + chunker + embed/cosine retrieval
  reusing the existing stack; Documents page (upload/paste, viewer, delete);
  top-k chunks injected into the system prompt.
- Vision chat: attach images, base64 into /api/chat.
- Voice I/O: Web Speech dictation + read-aloud (browser-native, no backend).
- Chat controls: stop, regenerate, edit-and-resend; num_ctx knob in Settings.
- Per-playbook model override.
- History polish: per-conversation + bulk ShareGPT export.
- Faster ncp start: UI up first, Ollama warms in the background, with a
  per-phase timing readout.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

tests/test_tools.py

@@ -0,0 +1,74 @@
+"""Tool-using playbook loop — the read-only MVP.
+
+Run from nexus-core/ with the Promethean venv active:  pytest -q
+
+Hermetic: a fake manager stands in for Ollama, so no network/model is needed.
+Guards the two pieces that would silently break the feature: the allowlist
+filter and the tool-call loop's terminate-on-content behaviour.
+"""
+import asyncio
+
+from synapse import tools
+from synapse.chat import _run_tool_loop
+
+
+def test_schemas_for_drops_unknown_names():
+    schemas = tools.schemas_for(["search_memory", "not_a_tool"])
+    names = [s["function"]["name"] for s in schemas]
+    assert names == ["search_memory"]
+    assert tools.schemas_for([]) == []
+
+
+async def _drain(gen):
+    return [s async for s in gen]
+
+
+class _FakeManager:
+    """Returns a tool_call on the first chat() call, plain content after."""
+    def __init__(self):
+        self.calls = 0
+
+    async def chat(self, **_):
+        self.calls += 1
+        if self.calls == 1:
+            return {
+                "role": "assistant",
+                "tool_calls": [
+                    {"function": {"name": "search_memory", "arguments": {"query": "gpu"}}}
+                ],
+            }
+        return {"role": "assistant", "content": "here is the answer"}
+
+
+def test_tool_loop_runs_tool_then_stops(monkeypatch):
+    async def fake_dispatch(name, args):
+        assert name == "search_memory"
+        assert args == {"query": "gpu"}
+        return '[{"section": "GPU", "text": "Vega 20 4GB"}]'
+
+    monkeypatch.setattr(tools, "dispatch", fake_dispatch)
+
+    messages = [{"role": "user", "content": "what gpu do i have?"}]
+    schemas = tools.schemas_for(["search_memory"])
+    statuses = asyncio.run(_drain(
+        _run_tool_loop(_FakeManager(), messages, "m", schemas, None, None)
+    ))
+
+    # one status sentinel per tool run
+    assert statuses == ["__status__search_memory"]
+    # messages mutated in place: user -> assistant(tool_calls) -> tool(result);
+    # the final content turn is NOT appended (the streaming turn regenerates it).
+    assert [m["role"] for m in messages] == ["user", "assistant", "tool"]
+    assert "Vega 20" in messages[-1]["content"]
+
+
+def test_tool_loop_degrades_when_model_returns_no_dict():
+    class _NoToolManager:
+        async def chat(self, **_):
+            return None  # model can't do tools / errored
+
+    messages = [{"role": "user", "content": "hi"}]
+    before = list(messages)
+    statuses = asyncio.run(_drain(_run_tool_loop(_NoToolManager(), messages, "m", [{}], None, None)))
+    assert statuses == []          # no tool ran
+    assert messages == before      # untouched -> falls back to a plain stream