ezmd 0.1.0rc1__tar.gz

ezmd-0.1.0rc1/.gitignore

@@ -0,0 +1,22 @@
+.venv/
+__pycache__/
+*.py[cod]
+.pytest_cache/
+.mypy_cache/
+.ruff_cache/
+.coverage
+coverage.xml
+htmlcov/
+dist/
+build/
+*.egg-info/
+node_modules/
+apps/web/dist/
+site/
+.env
+/data/
+/*.sqlite
+/*.db
+.DS_Store
+playwright-report/
+test-results/

ezmd-0.1.0rc1/PKG-INFO

@@ -0,0 +1,40 @@
+Metadata-Version: 2.5
+Name: ezmd
+Version: 0.1.0rc1
+Summary: Convert anything to LLM-ready Markdown.
+Project-URL: Homepage, https://github.com/larpey/ezmd
+Project-URL: Documentation, https://github.com/larpey/ezmd/tree/main/docs
+Project-URL: Repository, https://github.com/larpey/ezmd
+Project-URL: Issues, https://github.com/larpey/ezmd/issues
+Project-URL: Changelog, https://github.com/larpey/ezmd/blob/main/CHANGELOG.md
+Author: ezmd contributors
+License-Expression: Apache-2.0
+Requires-Python: >=3.12
+Requires-Dist: ezmd-converters==0.1.0rc1
+Requires-Dist: httpx>=0.28
+Requires-Dist: magika>=0.6
+Requires-Dist: markdown-it-py>=3.0
+Requires-Dist: pydantic>=2.9
+Requires-Dist: python-magic-bin>=0.4.14; sys_platform == 'win32'
+Requires-Dist: python-magic>=0.4.27; sys_platform != 'win32'
+Requires-Dist: pyyaml>=6.0
+Requires-Dist: rapidfuzz>=3.10
+Requires-Dist: rich>=13.9
+Requires-Dist: tiktoken>=0.8
+Requires-Dist: typer>=0.15
+Provides-Extra: 7z
+Requires-Dist: ezmd-converters[7z]; extra == '7z'
+Provides-Extra: all
+Requires-Dist: ezmd-converters[7z,data,docs]; extra == 'all'
+Requires-Dist: ezmd-mcp; extra == 'all'
+Provides-Extra: data
+Requires-Dist: ezmd-converters[data]; extra == 'data'
+Provides-Extra: docs
+Requires-Dist: ezmd-converters[docs]; extra == 'docs'
+Provides-Extra: mcp
+Requires-Dist: ezmd-mcp; extra == 'mcp'
+Provides-Extra: nonfree
+Requires-Dist: ezmd-converters[nonfree]; extra == 'nonfree'
+Description-Content-Type: text/markdown
+
+# core

ezmd-0.1.0rc1/README.md

@@ -0,0 +1 @@
+# core

ezmd-0.1.0rc1/pyproject.toml

@@ -0,0 +1,53 @@
+[project]
+name = "ezmd"
+version = "0.1.0rc1"
+description = "Convert anything to LLM-ready Markdown."
+readme = "README.md"
+requires-python = ">=3.12"
+license = "Apache-2.0"
+authors = [{ name = "ezmd contributors" }]
+dependencies = [
+    "ezmd-converters==0.1.0rc1",
+    "pydantic>=2.9",
+    "pyyaml>=6.0",
+    "tiktoken>=0.8",
+    "magika>=0.6",
+    "python-magic>=0.4.27; sys_platform != 'win32'",
+    "python-magic-bin>=0.4.14; sys_platform == 'win32'",
+    "rapidfuzz>=3.10",
+    "typer>=0.15",
+    "rich>=13.9",
+    "httpx>=0.28",
+    "markdown-it-py>=3.0",
+]
+
+
+[project.urls]
+Homepage = "https://github.com/larpey/ezmd"
+Documentation = "https://github.com/larpey/ezmd/tree/main/docs"
+Repository = "https://github.com/larpey/ezmd"
+Issues = "https://github.com/larpey/ezmd/issues"
+Changelog = "https://github.com/larpey/ezmd/blob/main/CHANGELOG.md"
+[project.optional-dependencies]
+# docs/spec/part4.md 4.3.3: users install `ezmd[<extra>]`; the converter extras live on ezmd-converters
+# (D-0003), so ezmd forwards them. media, ocr, web and fetch arrive with their phases; server needs ezmd-api,
+# which is not published yet.
+docs = ["ezmd-converters[docs]"]
+data = ["ezmd-converters[data]"]
+7z = ["ezmd-converters[7z]"]
+mcp = ["ezmd-mcp"]
+all = ["ezmd-converters[docs,data,7z]", "ezmd-mcp"]
+# Forbidden licenses (GPL extract-msg); never part of `all` (docs/spec/part4.md 4.3.3).
+nonfree = ["ezmd-converters[nonfree]"]
+
+[project.scripts]
+ezmd = "ezmd.cli:app"
+ezmd-score = "ezmd.testing.cli:score"
+ezmd-golden = "ezmd.testing.cli:golden"
+
+[build-system]
+requires = ["hatchling>=1.26"]
+build-backend = "hatchling.build"
+
+[tool.hatch.build.targets.wheel]
+packages = ["src/ezmd"]

ezmd-0.1.0rc1/src/ezmd/__init__.py

@@ -0,0 +1,48 @@
+"""ezmd: convert anything to LLM-ready Markdown.
+
+    >>> import ezmd
+    >>> result = ezmd.convert("report.pdf", profile="compact")
+    >>> print(result.markdown)
+
+Names are resolved lazily so `import ezmd` stays fast (no engines, no renderer at import time).
+"""
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING, Any
+
+__version__ = "0.1.0rc1"
+
+__all__ = [
+    "Options",
+    "Progress",
+    "Result",
+    "__version__",
+    "capabilities",
+    "convert",
+    "convert_async",
+    "convert_many",
+    "register_converter",
+    "unload_models",
+]
+
+if TYPE_CHECKING:
+    from ezmd.library import (
+        Options,
+        Progress,
+        Result,
+        capabilities,
+        convert,
+        convert_async,
+        convert_many,
+        register_converter,
+        unload_models,
+    )
+
+
+def __getattr__(name: str) -> Any:
+    if name in __all__ and name != "__version__":
+        from ezmd import library
+
+        return getattr(library, name)
+    raise AttributeError(f"module 'ezmd' has no attribute {name!r}")

ezmd-0.1.0rc1/src/ezmd/builtin.py

@@ -0,0 +1,26 @@
+"""ezmd.builtin: registers the converters shipped in `ezmd-converters`.
+
+Built-in converters live in the separate `ezmd_converters` package (docs/spec/part1.md section 1.3
+layout). They are registered here with source "builtin" rather than through entry points so that
+`ezmd` always resolves them first and a broken third-party plugin cannot shadow them (D-0003).
+"""
+
+from __future__ import annotations
+
+import logging
+from typing import TYPE_CHECKING
+
+if TYPE_CHECKING:
+    from ezmd.registry import ConverterRegistry
+
+log = logging.getLogger(__name__)
+
+
+def register_builtins(reg: ConverterRegistry) -> None:
+    try:
+        from ezmd_converters import builtin_converters
+    except ImportError as e:
+        log.warning("ezmd-converters is not installed; no built-in converters: %s", e)
+        return
+    for conv in builtin_converters():
+        reg.register(conv, source="builtin")  # Unavailable entries are recorded with their reason

ezmd-0.1.0rc1/src/ezmd/chains.py

@@ -0,0 +1,27 @@
+"""ezmd.chains: default fallback chains per mime type (docs/spec/part1.md section 5.3).
+
+Exact keys are matched before wildcard keys (`audio/*`). Each converter family declares the chains for
+the mimes it owns in `ezmd_converters.<family>.CHAINS` (D-0020); `DEFAULT_CHAINS` holds only chains
+that span families or exist without ezmd-converters installed.
+"""
+
+from __future__ import annotations
+
+import logging
+
+log = logging.getLogger(__name__)
+
+DEFAULT_CHAINS: dict[str, list[str]] = {}
+
+
+def default_chains() -> dict[str, list[str]]:
+    chains = dict(DEFAULT_CHAINS)
+    try:
+        from ezmd_converters import builtin_chains
+    except ImportError:
+        return chains
+    for mime, ids in builtin_chains().items():
+        if mime in chains:
+            log.warning("family chain for %s overrides the core default", mime)
+        chains[mime] = ids
+    return chains

ezmd-0.1.0rc1/src/ezmd/cli/__init__.py

@@ -0,0 +1,44 @@
+"""The `ezmd` command line (docs/spec/part4.md 4.2).
+
+Commands: convert, batch, doctor, capabilities, detect, version, serve. Every command is a thin client
+over the library (`ezmd.convert`, `ezmd.capabilities`) or, with `--remote`, the REST API.
+
+Exit codes (part4 4.2.2): 0 success; 1 generic failure; 2 bad arguments or missing dependency; 3 partial
+success (a usable result that carries an error-severity warning, D-0017; or a batch with failures under
+--continue-on-error); 4 fetch blocked by platform; 5 input too large; 6 unsupported type; 130 interrupted.
+"""
+
+from __future__ import annotations
+
+from ezmd.cli.app import app
+from ezmd.cli.exitcodes import (
+    EXIT_ARGS,
+    EXIT_BLOCKED,
+    EXIT_FAIL,
+    EXIT_INTERRUPTED,
+    EXIT_OK,
+    EXIT_PARTIAL,
+    EXIT_TOO_LARGE,
+    EXIT_UNSUPPORTED,
+)
+
+__all__ = [
+    "EXIT_ARGS",
+    "EXIT_BLOCKED",
+    "EXIT_FAIL",
+    "EXIT_INTERRUPTED",
+    "EXIT_OK",
+    "EXIT_PARTIAL",
+    "EXIT_TOO_LARGE",
+    "EXIT_UNSUPPORTED",
+    "app",
+    "main",
+]
+
+
+def main() -> None:  # pragma: no cover
+    app()
+
+
+if __name__ == "__main__":  # pragma: no cover
+    main()

ezmd-0.1.0rc1/src/ezmd/cli/app.py

@@ -0,0 +1,47 @@
+"""The typer application: command registration and logging setup (docs/spec/part4.md 4.2)."""
+
+from __future__ import annotations
+
+import logging
+
+import typer
+
+from ezmd.cli.batch import batch
+from ezmd.cli.convert import convert
+from ezmd.cli.doctor import doctor
+from ezmd.cli.info import capabilities, detect, serve, version
+from ezmd.cli.shadow import shadow_run
+
+__all__ = ["app"]
+
+app = typer.Typer(
+    add_completion=True,
+    no_args_is_help=True,
+    help="Convert anything to LLM-ready Markdown.",
+    context_settings={"help_option_names": ["-h", "--help"]},
+    pretty_exceptions_enable=False,
+)
+
+
+@app.callback()
+def _setup() -> None:
+    """Convert anything to LLM-ready Markdown."""
+    logger = logging.getLogger("ezmd")
+    if not any(getattr(h, "_ezmd_cli", False) for h in logger.handlers):
+        from rich.logging import RichHandler
+
+        from ezmd.cli.output import err_console
+
+        handler = RichHandler(console=err_console(), show_time=False, show_path=False, level=logging.WARNING)
+        handler._ezmd_cli = True  # type: ignore[attr-defined]
+        logger.addHandler(handler)
+
+
+app.command()(convert)
+app.command()(batch)
+app.command()(doctor)
+app.command()(capabilities)
+app.command()(detect)
+app.command()(version)
+app.command()(serve)
+app.command(name="shadow-run")(shadow_run)

ezmd-0.1.0rc1/src/ezmd/cli/batch.py

@@ -0,0 +1,366 @@
+"""`ezmd batch`: convert a directory or glob into a mirrored output tree, idempotently.
+
+Idempotency: before converting, each input's SHA-256 is compared with the previous run's record, taken
+from the previous manifest line for that input (`sha256`, plus matching profile and format) or, failing
+that, from the output's sidecar (`frontmatter.source_hash` and `profile`). A match with the output still
+present is reported as `skipped`. Conversion goes through `ezmd.convert`, in-process for one worker and
+in a process pool otherwise.
+"""
+
+from __future__ import annotations
+
+import glob as globlib
+import hashlib
+import json
+import multiprocessing
+import os
+import time
+from concurrent.futures import FIRST_COMPLETED, Future, ProcessPoolExecutor, wait
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Annotated, Any
+
+import typer
+
+from ezmd.cli.config import SUPPORTED_FORMATS, ConfigError, check_format, check_profile
+from ezmd.cli.convert import build_options, emit_json, fail, load_config_or_exit
+from ezmd.cli.exitcodes import EXIT_FAIL, EXIT_INTERRUPTED, EXIT_OK, EXIT_PARTIAL, classify
+from ezmd.cli.output import EXTENSIONS, err_console, outcome_from_result, sidecar_path, write_outcome
+
+__all__ = ["Job", "batch", "convert_job", "default_workers", "plan"]
+
+_GLOB_CHARS = set("*?[")
+_HASH_CHUNK = 1024 * 1024
+
+
+def default_workers() -> int:
+    """CPU count minus one, at least 1, at most 8 (Part 4 4.2.2)."""
+    return max(1, min((os.cpu_count() or 2) - 1, 8))
+
+
+@dataclass(frozen=True, slots=True)
+class Job:
+    """One input to convert. Plain data so it pickles into worker processes."""
+
+    src: str
+    out: str
+    profile: str
+    format: str
+    options: dict[str, object]
+    sidecar: bool
+    sha256: str
+
+
+def file_sha256(path: Path) -> str:
+    h = hashlib.sha256()
+    with path.open("rb") as f:
+        while chunk := f.read(_HASH_CHUNK):
+            h.update(chunk)
+    return h.hexdigest()
+
+
+def _hidden(rel: Path) -> bool:
+    return any(part.startswith(".") for part in rel.parts)
+
+
+def _within(path: Path, root: Path) -> bool:
+    try:
+        path.relative_to(root)
+    except ValueError:
+        return False
+    return True
+
+
+def discover(
+    target: str, recursive: bool, out_dir: Path, manifest: Path, exclude: frozenset[str] = frozenset()
+) -> tuple[Path, list[Path]]:
+    """Input files (sorted) and the base directory the output tree mirrors. Dotfiles, the output tree,
+    the manifest, sidecar files, and `exclude` (previous outputs, as resolved posix paths) are never inputs."""
+    out_root = out_dir.resolve()
+    if any(c in target for c in _GLOB_CHARS):
+        parts = Path(target).parts
+        fixed = list(parts[: next(i for i, p in enumerate(parts) if any(c in p for c in _GLOB_CHARS))])
+        base = Path(*fixed) if fixed else Path(".")
+        found = [Path(p) for p in globlib.glob(target, recursive=recursive)]
+    else:
+        root = Path(target)
+        if root.is_file():
+            return root.parent, [root]
+        if not root.is_dir():
+            raise FileNotFoundError(target)
+        base = root
+        found = list(root.rglob("*") if recursive else root.iterdir())
+    files: list[Path] = []
+    for p in found:
+        if not p.is_file():
+            continue
+        resolved = p.resolve()
+        rel = resolved.relative_to(base.resolve()) if _within(resolved, base.resolve()) else Path(p.name)
+        if _hidden(rel) or resolved == manifest.resolve() or p.name.endswith(".ezmd.json"):
+            continue
+        if resolved.as_posix() in exclude:
+            continue
+        if _within(resolved, out_root) and out_root != base.resolve():
+            continue
+        files.append(p)
+    return base, sorted(files, key=lambda q: q.as_posix())
+
+
+def _key(path: Path | str) -> str:
+    return Path(path).resolve().as_posix()
+
+
+def read_manifest(path: Path) -> dict[str, dict[str, Any]]:
+    """Previous manifest lines keyed by resolved input path. A corrupt line is ignored."""
+    prev: dict[str, dict[str, Any]] = {}
+    try:
+        lines = path.read_text(encoding="utf-8").splitlines()
+    except OSError:
+        return prev
+    for line in lines:
+        try:
+            rec = json.loads(line)
+        except ValueError:
+            rec = None
+        if isinstance(rec, dict) and isinstance(rec.get("path"), str):
+            prev[_key(rec["path"])] = rec
+    return prev
+
+
+def _sidecar_matches(target: Path, sha: str, profile: str) -> bool:
+    side = sidecar_path(target)
+    if not side.is_file():
+        return False
+    try:
+        data = json.loads(side.read_text(encoding="utf-8"))
+    except (OSError, ValueError):
+        return False
+    front = data.get("frontmatter") if isinstance(data, dict) else None
+    if not isinstance(front, dict):
+        return False
+    return front.get("source_hash") == f"sha256:{sha}" and data.get("profile") == profile
+
+
+def unchanged(job: Job, prev: dict[str, Any] | None) -> bool:
+    target = Path(job.out)
+    if not target.is_file():
+        return False
+    if (
+        prev is not None
+        and prev.get("sha256") == job.sha256
+        and prev.get("profile") == job.profile
+        and prev.get("format") == job.format
+        and prev.get("status") in ("converted", "partial", "skipped")
+    ):
+        return True
+    return job.format != "json" and _sidecar_matches(target, job.sha256, job.profile)
+
+
+def plan(base: Path, files: list[Path], out_dir: Path, fmt: str) -> list[tuple[Path, Path]]:
+    """(input, output) pairs mirroring the input tree. Inputs whose stems collide in one directory
+    (a.txt and a.md) keep their full name (a.txt.md, a.md.md) so no output overwrites another."""
+    ext = EXTENSIONS[fmt]
+    rels = []
+    for f in files:
+        resolved = f.resolve()
+        b = base.resolve()
+        rels.append(resolved.relative_to(b) if _within(resolved, b) else Path(f.name))
+    stems: dict[Path, int] = {}
+    for r in rels:
+        stems[r.parent / r.stem] = stems.get(r.parent / r.stem, 0) + 1
+    pairs = []
+    for f, r in zip(files, rels, strict=True):
+        name = r.stem if stems[r.parent / r.stem] == 1 else r.name
+        pairs.append((f, out_dir / r.parent / f"{name}{ext}"))
+    return pairs
+
+
+def convert_job(job: Job) -> dict[str, Any]:
+    """Convert one input and write it; never raises (runs in worker processes). Returns a manifest record."""
+    import ezmd.library as lib
+
+    start = time.perf_counter()
+    record: dict[str, Any] = {
+        "path": job.src,
+        "out": job.out,
+        "sha256": job.sha256,
+        "profile": job.profile,
+        "format": job.format,
+    }
+    try:
+        result = lib.convert(job.src, profile=job.profile, options=lib.Options(**job.options))
+        outcome = outcome_from_result(result, source=job.src, profile=job.profile, fmt=job.format)
+        write_outcome(outcome, Path(job.out), sidecar=job.sidecar)
+        record.update(
+            status="partial" if outcome.status == "partial" else "converted",
+            warnings=outcome.warnings,
+            tokens=outcome.tokens,
+        )
+    except Exception as e:
+        f = classify(e)
+        record.update(
+            status="failed",
+            out=None,
+            warnings=[{"kind": f.warning_kind, "severity": "error", "message": f.message}],
+            tokens=0,
+            error={"code": f.code, "message": f.message, "exit_code": f.exit_code},
+        )
+    record["seconds"] = round(time.perf_counter() - start, 3)
+    return record
+
+
+class _Run:
+    """Collects records, streams them to the manifest, and drives the progress bar."""
+
+    def __init__(self, manifest: Path, show: bool, total: int) -> None:
+        manifest.parent.mkdir(parents=True, exist_ok=True)
+        self.fh = manifest.open("w", encoding="utf-8", newline="\n")
+        self.records: list[dict[str, Any]] = []
+        self.bar: Any = None
+        self.task: Any = None
+        if show and total and err_console().is_terminal:
+            from rich.progress import BarColumn, MofNCompleteColumn, Progress, TextColumn, TimeElapsedColumn
+
+            cols = (TextColumn("Converting"), BarColumn(), MofNCompleteColumn(), TimeElapsedColumn())
+            self.bar = Progress(*cols, console=err_console(), transient=True)
+            self.bar.start()
+            self.task = self.bar.add_task("batch", total=total)
+
+    def add(self, rec: dict[str, Any], advance: bool = True) -> None:
+        self.records.append(rec)
+        self.fh.write(json.dumps(rec, ensure_ascii=False) + "\n")
+        self.fh.flush()
+        if advance and self.bar is not None:
+            self.bar.advance(self.task)
+
+    def close(self) -> None:
+        if self.bar is not None:
+            self.bar.stop()
+        self.fh.close()
+
+
+def _execute(jobs: list[Job], workers: int, keep_going: bool, run: _Run) -> bool:
+    """Run jobs; returns False when stopped early by a failure (without --continue-on-error)."""
+    if workers <= 1 or len(jobs) <= 1:
+        for i, job in enumerate(jobs):
+            rec = convert_job(job)
+            run.add(rec)
+            if rec["status"] == "failed" and not keep_going:
+                for rest in jobs[i + 1 :]:
+                    run.add(_cancelled(rest), advance=False)
+                return False
+        return True
+    # spawn, not fork: forking a process that already runs threads (Magika, tokenizers) can deadlock
+    with ProcessPoolExecutor(max_workers=workers, mp_context=multiprocessing.get_context("spawn")) as pool:
+        pending: dict[Future[dict[str, Any]], Job] = {pool.submit(convert_job, j): j for j in jobs}
+        stopped = False
+        while pending:
+            done, _ = wait(pending, return_when=FIRST_COMPLETED)
+            for fut in done:
+                job = pending.pop(fut)
+                if fut.cancelled():
+                    run.add(_cancelled(job), advance=False)
+                    continue
+                rec = fut.result()
+                run.add(rec)
+                if rec["status"] == "failed" and not keep_going and not stopped:
+                    stopped = True
+                    for other in pending:
+                        other.cancel()
+        return not stopped
+
+
+def _cancelled(job: Job) -> dict[str, Any]:
+    return {"path": job.src, "status": "cancelled", "out": None, "sha256": job.sha256, "profile": job.profile,
+            "format": job.format, "warnings": [], "seconds": 0.0, "tokens": 0}  # fmt: skip
+
+
+def _summary(records: list[dict[str, Any]]) -> dict[str, int]:
+    counts = {"converted": 0, "skipped": 0, "failed": 0, "cancelled": 0}
+    for r in records:
+        status = "converted" if r["status"] == "partial" else str(r["status"])
+        counts[status] = counts.get(status, 0) + 1
+    counts["total_tokens"] = sum(int(r.get("tokens") or 0) for r in records)
+    counts["total"] = len(records)
+    return counts
+
+
+def _print_table(summary: dict[str, int], records: list[dict[str, Any]], manifest: Path) -> None:
+    from rich.table import Table
+
+    con = err_console()
+    for r in records:
+        if r["status"] == "failed":
+            con.print(f"FAIL {r['path']}: {r['error']['message']}", markup=False, highlight=False)
+    t = Table("converted", "skipped", "failed", "total tokens", title="ezmd batch")
+    t.add_row(*(str(summary[k]) for k in ("converted", "skipped", "failed", "total_tokens")))
+    con.print(t)
+    con.print(f"manifest: {manifest}", markup=False, highlight=False)
+
+
+def batch(
+    inputs: Annotated[str, typer.Argument(help="Directory or glob (quote globs).")],
+    out: Annotated[Path | None, typer.Option("--out", "-o", help="Output directory (mirrors the input tree).")] = None,
+    recursive: Annotated[bool, typer.Option("--recursive", "-r", help="Descend into subdirectories.")] = False,
+    workers: Annotated[int | None, typer.Option("--workers", "-w", min=1, help="Worker processes.")] = None,
+    profile: Annotated[str | None, typer.Option("--profile", "-p", help="full | compact | rag | agent")] = None,
+    fmt: Annotated[str | None, typer.Option("--format", "-f", help="md | txt | json")] = None,
+    keep_going: Annotated[bool, typer.Option("--continue-on-error", help="Keep going after a failure.")] = False,
+    manifest: Annotated[Path | None, typer.Option("--manifest", help="[default: <out>/manifest.jsonl]")] = None,
+    sidecar: Annotated[bool | None, typer.Option("--sidecar/--no-sidecar", help="Write sidecars.")] = None,
+    opt: Annotated[list[str] | None, typer.Option("--opt", help="key=value converter or profile option.")] = None,
+    as_json: Annotated[bool, typer.Option("--json", help="Summary JSON on stdout.")] = False,
+    quiet: Annotated[bool, typer.Option("--quiet", "-q", help="No progress or table on stderr.")] = False,
+) -> None:
+    """Convert many files; unchanged inputs are skipped on re-runs."""
+    cfg = load_config_or_exit(as_json=as_json, quiet=quiet)
+    try:
+        chosen_fmt = check_format(fmt or cfg.format, "--format")
+        if chosen_fmt not in SUPPORTED_FORMATS:
+            raise ConfigError(f"--format {chosen_fmt} is not yet supported; use md, txt, or json")
+        chosen_profile = check_profile(profile or cfg.profile or "full", "--profile")
+        options = build_options(cfg, opt or [], None, None)
+        import ezmd.library as lib
+
+        lib.Options(**options)  # validate once, before any work
+        out_dir = out if out is not None else (Path(cfg.out_dir) if cfg.out_dir else None)
+        if out_dir is None:
+            raise ValueError("--out is required (or set defaults.out_dir in the config file)")
+        manifest_path = manifest or out_dir / "manifest.jsonl"
+        prev = read_manifest(manifest_path)
+        outputs = frozenset(_key(r["out"]) for r in prev.values() if isinstance(r.get("out"), str))
+        base, files = discover(inputs, recursive, out_dir, manifest_path, outputs)
+    except FileNotFoundError as e:
+        raise fail(f"no such file or directory: {inputs}", as_json=as_json, source=inputs) from e
+    except ValueError as e:
+        raise fail(str(e), as_json=as_json, source=inputs) from e
+    write_sidecar = cfg.sidecar if sidecar is None else sidecar
+    run = _Run(manifest_path, show=not quiet and not as_json, total=len(files))
+    finished = True
+    try:
+        todo: list[Job] = []
+        for src, target in plan(base, files, out_dir, chosen_fmt):
+            job = Job(str(src), str(target), chosen_profile, chosen_fmt, options, write_sidecar, file_sha256(src))
+            old = prev.get(_key(src))
+            if unchanged(job, old):
+                rec = {"path": job.src, "status": "skipped", "out": job.out, "sha256": job.sha256,
+                       "profile": job.profile, "format": job.format,
+                       "warnings": (old or {}).get("warnings", []), "seconds": 0.0,
+                       "tokens": int((old or {}).get("tokens") or 0)}  # fmt: skip
+                run.add(rec)
+            else:
+                todo.append(job)
+        finished = _execute(todo, workers or default_workers(), keep_going, run)
+    except KeyboardInterrupt:
+        run.close()
+        err_console().print("interrupted", markup=False)
+        raise typer.Exit(EXIT_INTERRUPTED) from None
+    run.close()
+    summary = _summary(run.records)
+    failed = summary["failed"] > 0
+    code = EXIT_OK if not failed else (EXIT_PARTIAL if keep_going and finished else EXIT_FAIL)
+    if as_json:
+        emit_json({**summary, "exit_code": code, "manifest": str(manifest_path), "files": run.records})
+    elif not quiet:
+        _print_table(summary, run.records, manifest_path)
+    raise typer.Exit(code)