local-knowledge 0.2.1__tar.gz → 0.2.3__tar.gz

{local_knowledge-0.2.1 → local_knowledge-0.2.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: local-knowledge
-Version: 0.2.1
+Version: 0.2.3
 Summary: Local file/folder indexer + grounded-Q&A MCP service — same engine as confluence-knowledge, no credentials needed
 License: MIT
 Project-URL: Repository, https://github.com/nkunal1992/agents
@@ -20,7 +20,7 @@ Requires-Dist: openpyxl>=3.1
 Requires-Dist: mcp<2,>=1.6
 Requires-Dist: pip-system-certs>=5.0
 Provides-Extra: models
-Requires-Dist: confluence-knowledge[models]>=0.5.1; extra == "models"
+Requires-Dist: confluence-knowledge[models]>=0.5.3; extra == "models"
 Provides-Extra: dev
 Requires-Dist: pytest>=8.0; extra == "dev"
 Requires-Dist: pytest-cov>=5.0; extra == "dev"

{local_knowledge-0.2.1 → local_knowledge-0.2.3}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "local-knowledge"
-version = "0.2.1"
+version = "0.2.3"
 description = "Local file/folder indexer + grounded-Q&A MCP service — same engine as confluence-knowledge, no credentials needed"
 readme = "README.md"
 requires-python = ">=3.11"
@@ -27,7 +27,7 @@ dependencies = [
 
 [project.optional-dependencies]
 models = [
-    "confluence-knowledge[models]>=0.5.1",
+    "confluence-knowledge[models]>=0.5.3",
 ]
 dev = [
     "pytest>=8.0",

{local_knowledge-0.2.1 → local_knowledge-0.2.3}/src/local_knowledge/cli_local.py

@@ -79,6 +79,9 @@ def main() -> None:
         import subprocess
         cmd = [sys.executable, "-m", "pip", "install", "-U",
                "local-knowledge[models]"]
+        # a running MCP server holds open files — quit the IDE first
+        print("note: quit the IDE (or stop the MCP server) before updating —"
+              " a running server can't be replaced cleanly", flush=True)
         if os.name == "nt":
             # this exe is the running process on Windows — pip can't replace
             # a file that's in use; hand the user the equivalent command

{local_knowledge-0.2.1 → local_knowledge-0.2.3}/src/local_knowledge/crawl_local.py

@@ -32,7 +32,7 @@ def resolve_root(path: str) -> Path:
     return p.resolve()
 
 
-from confluence_knowledge.retrieval import EMBED_MODEL_ID as _MODEL_ID
+from confluence_knowledge.retrieval import MODEL_MARKER as _MODEL_ID
 
 
 class LocalCrawler:
@@ -133,6 +133,19 @@ class LocalCrawler:
 
         queue: list[str] = job["queue"]
         errors: list[str] = job["errors"]
+        keep_failed: set[str] = set()  # transient read fails keep old content
+        # Mid-crawl model change: already-embedded pages live in the OLD
+        # vector space — finishing would publish mixed dims. Wipe + fail;
+        # start_crawl rebuilds under the new model.
+        stamped = self.index.model_id()
+        if stamped is not None and stamped != _MODEL_ID and queue:
+            self.index.clear_scope()
+            errors = errors + [
+                (f"model changed mid-crawl ({stamped} → {_MODEL_ID}) — "
+                 "index cleared; run start_crawl")]
+            self.index.save_job(job_id, "error", [], errors)
+            return {"job_id": job_id, "state": "error",
+                    "remaining": 0, "errors": errors}
         seen_raw = self.index.meta_get(f"seen:{job_id}")
         seen: set[str] = set(json.loads(seen_raw)) if seen_raw else set()
         levels_raw = self.index.meta_get(f"levels:{job_id}")
@@ -174,6 +187,7 @@ class LocalCrawler:
                 page, chunks = self._page(path)
             except SetupError as e:
                 errors.append(str(e))
+                keep_failed.add(self._rel(path))
                 continue
             rel = page["page_id"]
             if rel in seen:
@@ -196,7 +210,7 @@ class LocalCrawler:
         # persist final seen/levels/warns too — the in-loop checkpoint runs
         # before each item is processed, so the last item would undercount
         self._checkpoint(job_id, queue, errors, seen, levels, warns)
-        self.index.publish(seen)
+        self.index.publish(seen | keep_failed)
         self.index.set_model_id(_MODEL_ID)
         self.index.save_job(job_id, "done" if not queue else "running",
                             queue, errors)

{local_knowledge-0.2.1 → local_knowledge-0.2.3}/src/local_knowledge/server_local.py

@@ -14,7 +14,9 @@ import threading
 from confluence_knowledge.client import AccessError, SecurityException
 from confluence_knowledge.index import Index
 from confluence_knowledge.retrieval import (
-    EMBED_MODEL_ID,
+    MODEL_MARKER as EMBED_MODEL_ID,
+)
+from confluence_knowledge.retrieval import (
     LocalEmbedder,
     LocalReranker,
     LocalSparseEmbedder,
@@ -30,7 +32,7 @@ from .settings_local import APP_NAME, LocalProfile, home, index_root, workdir
 mcp = FastMCP("local-knowledge")
 
 _scopes: dict[str, dict] = {}
-_evidence_sets: dict[str, list[dict]] = {}
+_evidence_sets: dict[str, dict] = {}  # set_id → {scope, ev}
 _indexes: dict[str, Index] = {}
 _EMBEDDER = None
 _RERANKER = None
@@ -267,7 +269,10 @@ def search(scope_handle: str, question: str) -> dict:
                         "— run start_crawl once to rebuild it, then search again"}
     ev = retrieve(idx, question, _embedder(), _reranker(), sparse=_sparse())
     set_id = f"ev:{len(_evidence_sets)}:{s['key'][:8]}"
-    _evidence_sets[set_id] = ev
+    # bound the session cache — sets are never reused after the turn ends
+    while len(_evidence_sets) >= 32:
+        _evidence_sets.pop(next(iter(_evidence_sets)))
+    _evidence_sets[set_id] = {"scope": s["key"], "ev": ev}
     return {
         "evidence_set": set_id,
         "coverage": _coverage(idx),
@@ -275,7 +280,9 @@ def search(scope_handle: str, question: str) -> dict:
         "evidence": [
             {"chunk_id": e["chunk_id"], "page_title": e["page_title"],
              "page_url": e["page_url"], "heading": e["heading"],
-             "score": e.get("rerank_score"), "text": e["text"][:2000]}
+             "score": e.get("rerank_score"), "text": e["text"][:12000],
+             "low_confidence": e.get("low_confidence", False),
+             "truncated": len(e["text"]) > 12000}
             for e in ev
         ],
     }
@@ -301,12 +308,12 @@ def get_evidence(scope_handle: str, chunk_ids: list[str]) -> dict:
 @mcp.tool()
 def verify_citations(scope_handle: str, evidence_set: str,
                      mappings: list[dict]) -> dict:
-    _scope(scope_handle)
-    ev = _evidence_sets.get(evidence_set)
-    if ev is None or not evidence_set.endswith(_scopes[scope_handle]["key"][:8]):
+    s = _scope(scope_handle)
+    rec = _evidence_sets.get(evidence_set)
+    if rec is None or rec["scope"] != s["key"]:
         raise SecurityException(
             "unknown or cross-scope evidence set — run search again and cite its evidence_set")
-    return check_citations(ev, mappings)
+    return check_citations(rec["ev"], mappings)
 
 
 def main() -> None:

{local_knowledge-0.2.1 → local_knowledge-0.2.3}/src/local_knowledge.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: local-knowledge
-Version: 0.2.1
+Version: 0.2.3
 Summary: Local file/folder indexer + grounded-Q&A MCP service — same engine as confluence-knowledge, no credentials needed
 License: MIT
 Project-URL: Repository, https://github.com/nkunal1992/agents
@@ -20,7 +20,7 @@ Requires-Dist: openpyxl>=3.1
 Requires-Dist: mcp<2,>=1.6
 Requires-Dist: pip-system-certs>=5.0
 Provides-Extra: models
-Requires-Dist: confluence-knowledge[models]>=0.5.1; extra == "models"
+Requires-Dist: confluence-knowledge[models]>=0.5.3; extra == "models"
 Provides-Extra: dev
 Requires-Dist: pytest>=8.0; extra == "dev"
 Requires-Dist: pytest-cov>=5.0; extra == "dev"

{local_knowledge-0.2.1 → local_knowledge-0.2.3}/src/local_knowledge.egg-info/requires.txt

@@ -12,4 +12,4 @@ pytest-cov>=5.0
 ruff>=0.6
 
 [models]
-confluence-knowledge[models]>=0.5.1
+confluence-knowledge[models]>=0.5.3