rlmgrep 0.1.3__py3-none-any.whl → 0.1.5__py3-none-any.whl

rlmgrep/__init__.py

@@ -1,2 +1,2 @@
 __all__ = ["__version__"]
-__version__ = "0.1.3"
+__version__ = "0.1.5"

rlmgrep/cli.py

@@ -72,11 +72,10 @@ def _parse_args(argv: list[str]) -> argparse.Namespace:
     parser.add_argument("pattern", nargs="?", help="Query string (interpreted by RLM)")
     parser.add_argument("paths", nargs="*", help="Files or directories")
 
-    parser.add_argument("-n", dest="line_numbers", action="store_true", help="Show line numbers")
-    parser.add_argument("-H", dest="with_filename", action="store_true", help="Always show filenames")
+    parser.add_argument("-n", dest="line_numbers", action="store_true", help="Show line numbers (default)")
     parser.add_argument("-r", dest="recursive", action="store_true", help="Recursive (directories are searched recursively by default)")
     parser.add_argument("--no-recursive", dest="recursive", action="store_false", help="Do not recurse directories")
-    parser.set_defaults(recursive=True)
+    parser.set_defaults(recursive=True, line_numbers=True)
 
     parser.add_argument("-C", dest="context", type=int, default=0, help="Context lines before/after")
     parser.add_argument("-A", dest="after", type=int, default=None, help="Context lines after")
@@ -239,6 +238,33 @@ def _build_markitdown(config: dict, warnings: list[str]):
             kwargs["base_url"] = api_base
         return OpenAI(**kwargs)
 
+    class _LiteLLMClient:
+        def __init__(self, api_key: str | None, api_base: str | None):
+            try:
+                import litellm  # type: ignore
+            except Exception as exc:
+                raise RuntimeError("litellm not available") from exc
+            self._litellm = litellm
+            self._api_key = api_key
+            self._api_base = api_base
+            self.chat = self._Chat(self)
+
+        class _Chat:
+            def __init__(self, parent):
+                self.completions = parent._Completions(parent)
+
+        class _Completions:
+            def __init__(self, parent):
+                self._parent = parent
+
+            def create(self, model: str, messages):
+                kwargs = {"model": model, "messages": messages}
+                if self._parent._api_key:
+                    kwargs["api_key"] = self._parent._api_key
+                if self._parent._api_base:
+                    kwargs["api_base"] = self._parent._api_base
+                return self._parent._litellm.completion(**kwargs)
+
     llm_client = None
     if enable_images:
         if not llm_model:
@@ -246,18 +272,27 @@ def _build_markitdown(config: dict, warnings: list[str]):
                 "markitdown_enable_images set but markitdown_image_llm_model missing; skipping images"
             )
             enable_images = False
-        elif llm_provider != "openai":
-            warnings.append(
-                f"markitdown image LLM provider '{llm_provider}' not supported; skipping images"
-            )
-            enable_images = False
         else:
-            llm_client = _openai_client(
-                llm_api_key,
-                llm_api_base,
-                "openai package missing; skipping image conversion",
-            )
-            if llm_client is None:
+            if llm_provider == "openai":
+                llm_client = _openai_client(
+                    llm_api_key,
+                    llm_api_base,
+                    "openai package missing; skipping image conversion",
+                )
+                if llm_client is None:
+                    enable_images = False
+            elif llm_provider in {"gemini", "anthropic"}:
+                try:
+                    llm_client = _LiteLLMClient(llm_api_key, llm_api_base)
+                except RuntimeError:
+                    warnings.append(
+                        "litellm not available; skipping image conversion"
+                    )
+                    enable_images = False
+            else:
+                warnings.append(
+                    f"markitdown image LLM provider '{llm_provider}' not supported; skipping images"
+                )
                 enable_images = False
 
     md_kwargs: dict[str, object] = {"enable_plugins": False}
@@ -513,13 +548,16 @@ def main(argv: list[str] | None = None) -> int:
     before = args.before if args.before is not None else args.context
     after = args.after if args.after is not None else args.context
 
+    use_color = sys.stdout.isatty() and not os.getenv("NO_COLOR")
+
     output_lines = render_matches(
         files=files,
         matches=verified,
         show_line_numbers=args.line_numbers,
-        show_filename=args.with_filename,
         before=before,
         after=after,
+        use_color=use_color,
+        heading=True,
     )
 
     if args.answer:

rlmgrep/config.py

@@ -25,7 +25,7 @@ DEFAULT_CONFIG_TEXT = "\n".join(
         "# markitdown_image_llm_api_base = \"\"",
         "# markitdown_image_llm_prompt = \"\"",
         "# markitdown_enable_audio = false",
-        "# markitdown_audio_model = \"gpt-4o-mini-transcribe\"",
+        "# markitdown_audio_model = \"gpt-4o-mini-transcribe-2025-12-15\"",
         "# markitdown_audio_provider = \"openai\"",
         "# markitdown_audio_api_key = \"\"",
         "# markitdown_audio_api_base = \"\"",

rlmgrep/render.py

@@ -2,23 +2,37 @@ from __future__ import annotations
 
 from .ingest import FileRecord
 
+COLOR_RESET = "\x1b[0m"
+COLOR_PATH = "\x1b[35m"
+COLOR_LINE_NO = "\x1b[32m"
+
+
+def _colorize(text: str, color: str, use_color: bool) -> str:
+    if not use_color:
+        return text
+    return f"{color}{text}{COLOR_RESET}"
+
+
+def _format_heading(path: str, use_color: bool) -> str:
+    if not path.startswith((".", "/")):
+        path = f"./{path}"
+    return _colorize(path, COLOR_PATH, use_color)
+
 
 def _format_line(
-    path: str,
     line_no: int,
     text: str,
     is_match: bool,
-    show_filename: bool,
     show_line_numbers: bool,
+    use_color: bool,
+    heading: bool,
 ) -> str:
     delim = ":" if is_match else "-"
-    if show_filename and show_line_numbers:
-        return f"{path}{delim}{line_no}{delim}{text}"
-    if show_filename:
-        return f"{path}{delim}{text}"
-    if show_line_numbers:
-        return f"{line_no}{delim}{text}"
-    return text
+    if not show_line_numbers:
+        return text
+    prefix = _colorize(str(line_no), COLOR_LINE_NO, use_color)
+    sep = "\t" if heading else ""
+    return f"{prefix}{delim}{sep}{text}"
 
 
 def _merge_ranges(ranges: list[tuple[int, int]]) -> list[tuple[int, int]]:
@@ -39,18 +53,22 @@ def render_matches(
     files: dict[str, FileRecord],
     matches: dict[str, list[int]],
     show_line_numbers: bool,
-    show_filename: bool,
     before: int,
     after: int,
+    use_color: bool = False,
+    heading: bool = True,
 ) -> list[str]:
     output: list[str] = []
-    multiple_files = len(files) > 1
-    show_filename = show_filename or multiple_files
 
-    for path in sorted(matches.keys()):
+    paths = sorted(matches.keys())
+    for idx, path in enumerate(paths):
         record = files.get(path)
         if record is None:
             continue
+        if heading:
+            if idx > 0:
+                output.append("")
+            output.append(_format_heading(path, use_color))
         lines = record.lines
         page_map = record.page_map
         n_lines = len(lines)
@@ -65,12 +83,12 @@ def render_matches(
                         text = f"{text}\tpage={page_map[line_no - 1]}"
                     output.append(
                         _format_line(
-                            path,
                             line_no,
                             text,
                             True,
-                            show_filename,
                             show_line_numbers,
+                            use_color,
+                            heading,
                         )
                     )
             continue
@@ -90,12 +108,12 @@ def render_matches(
                 is_match = line_no in match_set
                 output.append(
                     _format_line(
-                        path,
                         line_no,
                         text,
                         is_match,
-                        show_filename,
                         show_line_numbers,
+                        use_color,
+                        heading,
                     )
                 )
             if idx < len(merged) - 1:

rlmgrep/rlm.py

@@ -24,6 +24,10 @@ class RLMGrepSignature(dspy.Signature):
     find relevant lines. Return all relevant matches you can find, avoid duplicates,
     and only use exact paths from the directory keys.
     Always read the ASCII file map first to orient yourself to the available paths.
+    Do not wrap code in backticks; only raw Python.
+    Do not import pandas or numpy; use built-ins only.
+
+    Files like "photo.jpg.md" or "audio.mp3.md" are LLM descriptions/transcriptions of images/audio that were originally in the directory but have been converted to md to make them searchable by you. 
     """
 
     directory: dict = dspy.InputField(
@@ -52,7 +56,21 @@ class RLMGrepSignature(dspy.Signature):
 
 class RLMGrepAnswerSignature(dspy.Signature):
     """
-    Same as RLMGrepSignature, but also return a narrative answer to the query.
+    You are the search engine for rlmgrep, a grep-shaped CLI for coding agents.
+    Inputs include a directory mapping of files (path -> full text), an ASCII file
+    map, and a user query string. Your output must be grep-printable matches as
+    (path, line) pairs that point to real lines in the provided texts.
+    The query may be natural language or a short pattern; interpret it freely to
+    find relevant lines. Return all relevant matches you can find, avoid duplicates,
+    and only use exact paths from the directory keys.
+    Always read the ASCII file map first to orient yourself to the available paths.
+    Do not wrap code in backticks; only raw Python.
+    Do not import pandas or numpy; use built-ins only.
+
+    In this mode you are also responsible for generating a narrative answer to the query based on the provided files.
+
+    Files like "photo.jpg.md" or "audio.mp3.md" are LLM descriptions/transcriptions of images/audio that were originally in the directory but have been converted to md to make them searchable by you.
+
     """
 
     directory: dict = dspy.InputField(

{rlmgrep-0.1.3.dist-info → rlmgrep-0.1.5.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: rlmgrep
-Version: 0.1.3
+Version: 0.1.5
 Summary: Grep-shaped CLI search powered by DSPy RLM
 Author: rlmgrep
 License: MIT
@@ -61,8 +61,7 @@ rlmgrep [options] "query" [paths...]
 
 Common options:
 
-- `-n` show line numbers
-- `-H` always show filenames
+- `-n` show line numbers (default)
 - `-C N` context lines before/after (grep-style)
 - `-A N` context lines after
 - `-B N` context lines before
@@ -109,10 +108,9 @@ rg -l "token" . | rlmgrep --stdin-files --answer "what does this token control?"
 ## Output contract (stable for agents)
 
 - Matches are written to stdout; warnings go to stderr.
-- Output uses grep-like prefixes:
-  - `path:line:text` for match lines when both `-H` and `-n` are enabled.
-  - `path-line-text` for context lines (note the `-` separator).
-  - If `-H` or `-n` are omitted, their parts are omitted.
+- Output uses rg-style headings by default:
+  - A file header line like `./path/to/file`
+  - Then `line:\ttext` for matches, `line-\ttext` for context lines
 - Line numbers are 1-based.
 - When context ranges are disjoint, a `--` line separates groups.
 - Exit codes:
@@ -120,7 +118,7 @@ rg -l "token" . | rlmgrep --stdin-files --answer "what does this token control?"
   - `1` = no matches
   - `2` = usage/config/error
 
-Agent tip: use `-n -H` and no context for parse-friendly output, then key off exit codes.
+Agent tip: use `-n` and no context for parse-friendly output, then key off exit codes.
 
 ## Regex-style queries (best effort)
 
@@ -160,7 +158,7 @@ file_hard_max = 1000
 # markitdown_image_llm_api_base = ""
 # markitdown_image_llm_prompt = ""
 # markitdown_enable_audio = false
-# markitdown_audio_model = "gpt-4o-mini-transcribe"
+# markitdown_audio_model = "gpt-4o-mini-transcribe-2025-12-15"
 # markitdown_audio_provider = "openai"
 # markitdown_audio_api_key = ""
 # markitdown_audio_api_base = ""
@@ -180,7 +178,7 @@ If more than one provider key is set and the model does not make the provider ob
 ## Non-text files (PDF, images, audio)
 
 - PDF files are parsed with `pypdf`. Each page gets a marker line `===== Page N =====`, and output lines include a `page=N` suffix.
-- Images and audio are converted via `markitdown` when enabled in config. For image/audio conversion, an `openai` Python client is required.
+- Images and audio are converted via `markitdown` when enabled in config. Image conversion supports `openai`, `anthropic`, and `gemini` providers; audio conversion currently supports `openai` only.
 - Converted image/audio text is cached in sidecar files named `<original>.<ext>.md` next to the original file and reused on subsequent runs.
 - Use `-a/--text` to force binary files to be read as text (UTF-8 with replacement).
 
@@ -188,7 +186,7 @@ If more than one provider key is set and the model does not make the provider ob
 
 - Prefer narrow corpora (globs/types) to reduce token usage.
 - Use `--max-llm-calls` to cap costs; combine with small `--max-iterations` for safety.
-- For reproducible parsing, use `-n -H` and avoid context (`-C/-A/-B`).
+- For reproducible parsing, use `-n` and avoid context (`-C/-A/-B`).
   
 ## Development
 

rlmgrep-0.1.5.dist-info/RECORD

@@ -0,0 +1,14 @@
+rlmgrep/__init__.py,sha256=nkO5AhSYqNqN7gEHOfOJe6qbc5OMva0ggSpwLNV5QR8,48
+rlmgrep/__main__.py,sha256=MHKZ_ae3fSLGTLUUMOx15fWdeOnJSHhq-zslRP5F5Lc,79
+rlmgrep/cli.py,sha256=qCN6Nvepb-HvpbvCSp-NAyThuZbML6k89ndzAqlg_-I,20246
+rlmgrep/config.py,sha256=u1iz-nI8dj-dZETbpIki3RQefHJEyi5oE5zE4_IR8kg,2399
+rlmgrep/file_map.py,sha256=x2Ri1wzK8_87GUorsAV01K_nYLZcv30yIquDeTCcdEw,876
+rlmgrep/ingest.py,sha256=uCz2el9B-RIT9umFo-gFEdAsmWPP1IJOArFFQY0D_1A,9127
+rlmgrep/interpreter.py,sha256=s_nMRxLlAU9C0JmUzUBW5NbVbuH67doVWF54K54STlA,2478
+rlmgrep/render.py,sha256=OYZy7BuJJe-KsDhEGAz6JA5RGd65ZInPWf9wLDJE0ag,3554
+rlmgrep/rlm.py,sha256=i3rCTp8OABByF60Un5gO7265gaW4spwU0OFKIz4surg,5750
+rlmgrep-0.1.5.dist-info/METADATA,sha256=aY7Fy6DgS7lmtxFzmt79ry2PebKB43744ncjlT9mZHM,6584
+rlmgrep-0.1.5.dist-info/WHEEL,sha256=wUyA8OaulRlbfwMtmQsvNngGrxQHAvkKcvRmdizlJi0,92
+rlmgrep-0.1.5.dist-info/entry_points.txt,sha256=UV6QkEbkwBO1JJ53mm84_n35tVyOczPvOQ14ga7vrCI,45
+rlmgrep-0.1.5.dist-info/top_level.txt,sha256=gTujSRsO58c80eN7aRH2cfe51FHxx8LJ1w1Y2YlHti0,8
+rlmgrep-0.1.5.dist-info/RECORD,,

rlmgrep-0.1.3.dist-info/RECORD

@@ -1,14 +0,0 @@
-rlmgrep/__init__.py,sha256=cVZBCfo6mJZvGFsStEhk2sSrk77IfDImcTVxgYIhNmY,48
-rlmgrep/__main__.py,sha256=MHKZ_ae3fSLGTLUUMOx15fWdeOnJSHhq-zslRP5F5Lc,79
-rlmgrep/cli.py,sha256=wR9zJAzkp8jl42zMHL19r4oCxGKfN6K72-JzmQlUS74,18768
-rlmgrep/config.py,sha256=A6VLuuXSgQ1vM207CP0G92Mg3et93dGSmkkLQ0IOfwk,2388
-rlmgrep/file_map.py,sha256=x2Ri1wzK8_87GUorsAV01K_nYLZcv30yIquDeTCcdEw,876
-rlmgrep/ingest.py,sha256=uCz2el9B-RIT9umFo-gFEdAsmWPP1IJOArFFQY0D_1A,9127
-rlmgrep/interpreter.py,sha256=s_nMRxLlAU9C0JmUzUBW5NbVbuH67doVWF54K54STlA,2478
-rlmgrep/render.py,sha256=w6KOfont2M7pQz_EEngTFMY5xJEE11N_ko8P9x5FdH8,3097
-rlmgrep/rlm.py,sha256=LZfkyWxjvtf8dwo5JxetKvvpBYeGKhajwHEVpCb2eo4,4474
-rlmgrep-0.1.3.dist-info/METADATA,sha256=RuGjNIucLiFErCBf4KnH4An7lhgUE5vLIT3WwtmCBEY,6615
-rlmgrep-0.1.3.dist-info/WHEEL,sha256=wUyA8OaulRlbfwMtmQsvNngGrxQHAvkKcvRmdizlJi0,92
-rlmgrep-0.1.3.dist-info/entry_points.txt,sha256=UV6QkEbkwBO1JJ53mm84_n35tVyOczPvOQ14ga7vrCI,45
-rlmgrep-0.1.3.dist-info/top_level.txt,sha256=gTujSRsO58c80eN7aRH2cfe51FHxx8LJ1w1Y2YlHti0,8
-rlmgrep-0.1.3.dist-info/RECORD,,