gemcode 0.3.88__tar.gz → 0.3.90__tar.gz

{gemcode-0.3.88/src/gemcode.egg-info → gemcode-0.3.90}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gemcode
-Version: 0.3.88
+Version: 0.3.90
 Summary: Local-first coding agent on Google Gemini + ADK
 Author: GemCode Contributors
 License:                                  Apache License

{gemcode-0.3.88 → gemcode-0.3.90}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "gemcode"
-version = "0.3.88"
+version = "0.3.90"
 description = "Local-first coding agent on Google Gemini + ADK"
 readme = "README.md"
 requires-python = ">=3.11"

{gemcode-0.3.88 → gemcode-0.3.90}/src/gemcode/invoke.py

@@ -175,10 +175,26 @@ async def run_turn(
       if attachment_paths:
         from gemcode.multimodal_input import build_user_content
 
+        # Optional HITL gate for local attachment reads/upload materialization.
+        # This is separate from workspace trust: the OS may still require
+        # user-granted "Files and Folders" permissions on first access.
+        # If approved once, we don't re-prompt for the rest of this session.
+        attach_allow = True
+        if cfg is not None:
+          attach_allow = bool(getattr(cfg, "interactive_permission_ask", False))
+          attach_allow = attach_allow and hasattr(sys.stdin, "isatty") and sys.stdin.isatty()
+          if attach_allow and not bool(getattr(cfg, "_attachments_allowed", False)):
+            attach_allow = _prompt_yes_no(
+              "Allow GemCode to read and upload the attached file(s) from disk? (y/n) "
+            )
+            if attach_allow:
+              object.__setattr__(cfg, "_attachments_allowed", True)
+        effective_attachments = attachment_paths if attach_allow else None
+
         root = cfg.project_root if cfg is not None else Path.cwd()
         current_message, attach_warn = build_user_content(
             prompt,
-            attachment_paths,
+            effective_attachments,
             project_root=root,
         )
         for w in attach_warn:

gemcode-0.3.90/src/gemcode/tools/veomem_tools.py

@@ -0,0 +1,252 @@
+from __future__ import annotations
+
+import os
+from pathlib import Path
+from typing import Any
+
+
+def _enabled() -> bool:
+  return os.environ.get("GEMCODE_VEOMEM", "").strip().lower() in ("1", "true", "yes", "on")
+
+
+def make_veomem_tools(cfg) -> list[Any]:
+  """
+  VeoMem recall tools (token-efficient, 3-step retrieval workflow).
+
+  Exposes a compact flow to the agent:
+  1) `veomem_search` returns a small index of matching observation IDs
+  2) `veomem_timeline` fetches compact context around a specific anchor ID
+  3) `veomem_get_observations` fetches full text for a small set of IDs
+  """
+  if not _enabled():
+    return []
+
+  project_root = Path(getattr(cfg, "project_root"))
+
+  try:
+    from veomem.store import get_observations as _get_observations
+    from veomem.store import search as _search
+    from veomem.store import timeline as _timeline
+  except Exception:
+    return []
+
+  def veomem_search(
+    query: str,
+    limit: int = 10,
+    kind: str | None = None,
+    wing: str | None = None,
+    room: str | None = None,
+  ) -> dict[str, Any]:
+    """
+    Search VeoMem for relevant prior tool observations.
+
+    Returns a small index of hits (IDs + snippets). Typical usage:
+    - call `veomem_timeline()` for one promising anchor
+    - then call `veomem_get_observations()` for a small set of IDs
+    """
+    if not isinstance(query, str) or not query.strip():
+      return {"error": "query must be a non-empty string"}
+    return _search(
+      project_root,
+      query=str(query),
+      limit=int(limit),
+      kind=str(kind) if kind else None,
+      wing=str(wing) if wing else None,
+      room=str(room) if room else None,
+    )
+
+  def veomem_timeline(
+    id: int,
+    window_ms: int = 120000,
+    limit: int = 40,
+  ) -> dict[str, Any]:
+    """
+    Fetch compact context around a specific VeoMem observation id.
+
+    This is the "timeline" step: it returns neighbors around an anchor.
+    The agent then decides which IDs to fetch in full.
+    """
+    return _timeline(
+      project_root,
+      observation_id=int(id),
+      window_ms=int(window_ms),
+      limit=int(limit),
+    )
+
+  def veomem_get_observations(
+    ids: str,
+    max_chars: int = 8000,
+  ) -> dict[str, Any]:
+    """
+    Fetch full observation text for selected VeoMem observation IDs.
+
+    ids:
+      Comma-separated list, e.g. "12,13,21"
+
+    The returned `text` is truncated to `max_chars` per observation.
+    """
+    if not isinstance(ids, str) or not ids.strip():
+      return {"error": "ids must be a non-empty comma-separated string"}
+
+    parts = [x.strip() for x in ids.split(",") if x.strip()]
+    id_list: list[int] = []
+    for p in parts:
+      try:
+        id_list.append(int(p))
+      except Exception:
+        pass
+
+    if not id_list:
+      return {"error": "no valid integer ids found"}
+
+    rows = _get_observations(project_root, ids=id_list)
+    out: list[dict[str, Any]] = []
+    for o in rows:
+      t = (o.text or "").strip().replace("\n", " ")
+      truncated = False
+      if len(t) > int(max_chars):
+        t = t[: int(max_chars)].rstrip() + "…"
+        truncated = True
+      out.append(
+        {
+          "id": int(o.id),
+          "kind": str(o.kind),
+          "title": str(o.title),
+          "wing": str(o.wing),
+          "room": str(o.room),
+          "text": t,
+          "truncated": truncated,
+        }
+      )
+
+    return {"ok": True, "results": out, "count": len(out)}
+
+  veomem_search.__name__ = "veomem_search"
+  veomem_timeline.__name__ = "veomem_timeline"
+  veomem_get_observations.__name__ = "veomem_get_observations"
+
+  return [veomem_search, veomem_timeline, veomem_get_observations]
+import os
+from pathlib import Path
+from typing import Any
+
+
+def _enabled() -> bool:
+  return os.environ.get("GEMCODE_VEOMEM", "").strip().lower() in ("1", "true", "yes", "on")
+
+
+def make_veomem_tools(cfg) -> list[Any]:
+  """
+  VeoMem recall tools (compact 3-step retrieval workflow).
+
+  Exposes a token-efficient 3-step flow to the agent:
+  1) `veomem_search` returns an index of matching observation IDs
+  2) `veomem_timeline` fetches compact context around a specific anchor ID
+  3) `veomem_get_observations` fetches full text for a small set of IDs
+  """
+  if not _enabled():
+    return []
+
+  project_root = Path(getattr(cfg, "project_root"))
+
+  try:
+    from veomem.store import get_observations as _get_observations
+    from veomem.store import search as _search
+    from veomem.store import timeline as _timeline
+  except Exception:
+    return []
+
+  def veomem_search(
+    query: str,
+    limit: int = 10,
+    kind: str | None = None,
+    wing: str | None = None,
+    room: str | None = None,
+  ) -> dict[str, Any]:
+    """
+    Search VeoMem for relevant prior tool observations (FTS5/BM25).
+
+    Returns a small "index" of hits (IDs + snippets). Prefer:
+    - call `veomem_timeline()` for one promising anchor
+    - then call `veomem_get_observations()` for a small set of IDs
+    """
+    if not isinstance(query, str) or not query.strip():
+      return {"error": "query must be a non-empty string"}
+    return _search(
+      project_root,
+      query=str(query),
+      limit=int(limit),
+      kind=str(kind) if kind else None,
+      wing=str(wing) if wing else None,
+      room=str(room) if room else None,
+    )
+
+  def veomem_timeline(
+    id: int,
+    window_ms: int = 120000,
+    limit: int = 40,
+  ) -> dict[str, Any]:
+    """
+    Fetch compact context around a specific VeoMem observation id.
+
+    This is the "timeline" step: it returns neighbors around an anchor
+    (usually the agent should then decide which IDs to fetch in full).
+    """
+    return _timeline(
+      project_root,
+      observation_id=int(id),
+      window_ms=int(window_ms),
+      limit=int(limit),
+    )
+
+  def veomem_get_observations(
+    ids: str,
+    max_chars: int = 8000,
+  ) -> dict[str, Any]:
+    """
+    Fetch full observation text for selected VeoMem observation IDs.
+
+    ids:
+      Comma-separated list, e.g. "12,13,21"
+
+    The returned `text` is truncated to `max_chars` per observation.
+    """
+    if not isinstance(ids, str) or not ids.strip():
+      return {"error": "ids must be a non-empty comma-separated string"}
+    parts = [x.strip() for x in ids.split(",") if x.strip()]
+    id_list: list[int] = []
+    for p in parts:
+      try:
+        id_list.append(int(p))
+      except Exception:
+        pass
+    if not id_list:
+      return {"error": "no valid integer ids found"}
+
+    rows = _get_observations(project_root, ids=id_list)
+    out: list[dict[str, Any]] = []
+    for o in rows:
+      t = (o.text or "").strip().replace("\n", " ")
+      truncated = False
+      if len(t) > int(max_chars):
+        t = t[: int(max_chars)].rstrip() + "…"
+        truncated = True
+      out.append(
+        {
+          "id": int(o.id),
+          "kind": str(o.kind),
+          "title": str(o.title),
+          "wing": str(o.wing),
+          "room": str(o.room),
+          "text": t,
+          "truncated": truncated,
+        }
+      )
+    return {"ok": True, "results": out, "count": len(out)}
+
+  veomem_search.__name__ = "veomem_search"
+  veomem_timeline.__name__ = "veomem_timeline"
+  veomem_get_observations.__name__ = "veomem_get_observations"
+
+  return [veomem_search, veomem_timeline, veomem_get_observations]
+

{gemcode-0.3.88 → gemcode-0.3.90/src/gemcode.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gemcode
-Version: 0.3.88
+Version: 0.3.90
 Summary: Local-first coding agent on Google Gemini + ADK
 Author: GemCode Contributors
 License:                                  Apache License

{gemcode-0.3.88 → gemcode-0.3.90}/src/gemcode.egg-info/SOURCES.txt

@@ -104,6 +104,7 @@ src/gemcode/tools/subtask.py
 src/gemcode/tools/tasks.py
 src/gemcode/tools/think.py
 src/gemcode/tools/todo.py
+src/gemcode/tools/veomem_tools.py
 src/gemcode/tools/web.py
 src/gemcode/tools/web_search.py
 src/gemcode/tui/input_handler.py