Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.
Python agentaichatgptdeepseekfine-tuninggemmaimage-generationllamallmllmsopenaipythonqwenreinforcement-learningself-hostedstable-diffusiontext-to-speechttsuiunsloth
# SPDX-License-Identifier: AGPL-3.0-only
# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
"""#9485: picking a knowledge base in a project chat must not drop the project's sources."""
import asyncio
import json
import math
import pytest
from core.rag import store, tool
from core.rag.chunking import Chunk
VOCAB = ["alpha", "bravo", "company", "customer", "thread"]
def _embed(text):
v = [float(text.lower().count(w)) for w in VOCAB]
n = math.sqrt(sum(x * x for x in v)) or 1.0
return [x / n for x in v]
@pytest.fixture
def bow_embeddings(monkeypatch):
from core.rag import embeddings
monkeypatch.setattr(
embeddings,
"encode",
lambda texts, *, model_name = None, normalize = True: [_embed(t) for t in texts],
)
monkeypatch.setattr(embeddings, "dim", lambda model_name = None: len(VOCAB))
def _add_doc(conn, scope, doc_id, filename, text):
store.create_document(conn, scope = scope, filename = filename, sha256 = doc_id, document_id = doc_id)
chunk = Chunk(
text = text,
token_count = len(text.split()),
page_number = None,
source_page_index = 0,
chunk_index = 0,
page_char_start = 0,
page_char_end = len(text),
)
store.add_chunks(conn, scope, doc_id, [chunk], [_embed(text)])
conn.execute("UPDATE documents SET status='completed', num_chunks=1 WHERE id=?", (doc_id,))
conn.commit()
def test_scope_resolution():
assert tool._resolve_scope("K", "T", "P") == ["kb_K", "project_P"]
assert tool._resolve_scope("K", "T", None) == "kb_K"
assert tool._resolve_scope(None, "T", "P") == ["project_P", "thread_T"]
# The conversation archive stays exclusive.
assert tool._resolve_scope("K", "T", "P", "C") == store.conversation_archive_scope("C")
def _seed(conn):
_add_doc(conn, "kb_K", "dk", "company.pdf", "alpha bravo company policy")
_add_doc(conn, "project_P", "dp", "customer.pdf", "alpha bravo customer notes")
_add_doc(conn, "thread_T", "dt", "thread.pdf", "alpha bravo thread upload")
def test_search_with_kb_also_reads_project(rag_conn, bow_embeddings):
_seed(rag_conn)
_, sources = tool.search_knowledge_base_with_sources(
query = "alpha bravo", scope_kb_id = "K", scope_thread_id = "T", scope_project_id = "P"
)
assert {s["filename"] for s in sources} == {"company.pdf", "customer.pdf"}
def test_autoinject_with_kb_also_reads_project(rag_conn, bow_embeddings, monkeypatch):
from core.inference import tools
from storage import rag_db
monkeypatch.setattr(rag_db, "RAG_AVAILABLE", True, raising = False)
monkeypatch.setattr(rag_db, "rag_available", lambda: True, raising = False)
_seed(rag_conn)
out = tools.build_rag_autoinject(
[{"role": "user", "content": "alpha bravo"}],
{
"kb_id": "K",
"thread_id": "T",
"project_id": "P",
"autoinject": True,
"autoinject_min_score": 0.0,
},
)
blob = json.dumps(out)
assert "company.pdf" in blob and "customer.pdf" in blob and "thread.pdf" not in blob
def test_roster_with_kb_also_names_project(rag_conn):
from routes import inference
_seed(rag_conn)
tools = [{"type": "function", "function": {"name": "search_knowledge_base"}}]
out = asyncio.run(
inference._apply_rag_nudge(
"", tools, rag_scope = {"kb_id": "K", "thread_id": "T", "project_id": "P"}
)
)
assert '"company.pdf"' in out and '"customer.pdf"' in out and '"thread.pdf"' not in out