diff --git a/ir/__init__.py b/ir/__init__.py
index 45af6e1..61ff3cc 100644
--- a/ir/__init__.py
+++ b/ir/__init__.py
@@ -27,6 +27,7 @@
 from __future__ import annotations
 
 from . import embed as _embed  # noqa: F401  (sets USE_TF=0 before transformers)
+from . import registry
 from .base import Artifact, IndexPlan, Record, SearchHit, Surface
 from .index import Corpus, build, open_corpus
 from .retrieve import search as _search
@@ -51,8 +52,19 @@
     "build",
     "open_corpus",
     "search",
+    "register",
+    "corpora",
+    "build_corpus",
 ]
 
+register = registry.register
+corpora = registry.registered
+
+
+def build_corpus(name, **kwargs):
+    """Build (or update) a registered/preset corpus by name; returns a Corpus."""
+    return build(registry.source_for(name), **kwargs)
+
 
 def search(corpus, query, **kwargs):
     """Search a :class:`~ir.index.Corpus`, or a corpus *name* (reopened lazily)."""
diff --git a/ir/__main__.py b/ir/__main__.py
new file mode 100644
index 0000000..9d73db7
--- /dev/null
+++ b/ir/__main__.py
@@ -0,0 +1,18 @@
+"""``python -m ir`` / ``ir`` CLI entry point (argh dispatch)."""
+
+from __future__ import annotations
+
+from .cli import COMMANDS
+
+
+def main():
+    """Dispatch the ``ir`` command-line interface."""
+    import argh
+
+    parser = argh.ArghParser()
+    argh.add_commands(parser, COMMANDS)
+    parser.dispatch()
+
+
+if __name__ == "__main__":
+    main()
diff --git a/ir/cli.py b/ir/cli.py
new file mode 100644
index 0000000..14faaa3
--- /dev/null
+++ b/ir/cli.py
@@ -0,0 +1,83 @@
+"""Command-line surface for ``ir`` (argh-dispatched).
+
+Commands operate on **named** corpora from the registry (see
+:mod:`ir.registry`)::
+
+    ir build skills                 # build/update the skills preset corpus
+    ir search skills "deploy app"   # query it
+    ir ls                           # list corpora + record counts
+    ir info packages                # config + stats for a corpus
+    ir register notes files --root ~/notes --pattern '.*\\.md$'
+    ir rm notes                     # unregister (keeps built data)
+"""
+
+from __future__ import annotations
+
+from . import registry
+from .index import build as _build
+from .index import open_corpus
+
+
+def ls():
+    """List registered corpora with their kind, embedder, and record count."""
+    entries = registry.registered()
+    if not entries:
+        return "No corpora registered. Try: ir build skills"
+    lines = []
+    for name, e in entries.items():
+        try:
+            count = len(open_corpus(name))
+        except Exception:
+            count = 0
+        lines.append(
+            f"{name:18} {e['kind']:10} {e.get('embedder', 'default'):10} records={count}"
+        )
+    return "\n".join(lines)
+
+
+def register(name, kind, *, root=None, pattern=None, embedder="default"):
+    """Register a named corpus. kind: skills | packages | reports | files."""
+    params = {}
+    if root:
+        params["root"] = root
+    if pattern:
+        params["pattern"] = pattern
+    registry.register(name, kind, embedder=embedder, **params)
+    return f"registered {name!r} (kind={kind}, embedder={embedder})"
+
+
+def build(name, *, embedder=None, full=True):
+    """Build or incrementally update a registered (or preset) corpus."""
+    source = registry.source_for(name)
+    corpus = _build(source, embedder=embedder, full=full)
+    return f"built {name!r}: {len(corpus)} records (embedder {corpus.embedder_id})"
+
+
+def search(name, query, *, k=10):
+    """Search a built corpus and print the top-k hits."""
+    corpus = open_corpus(name)
+    if len(corpus) == 0:
+        return f"corpus {name!r} is empty; build it first: ir build {name}"
+    lines = []
+    for h in corpus.search(query, k=k):
+        # artifact_id is the unique label across corpora (skill name[@parent],
+        # package name, or a report's relative path).
+        lines.append(f"{h.score:+.3f}  {h.artifact_id}  [{h.surface_kind}]")
+    return "\n".join(lines) or "(no matches)"
+
+
+def info(name):
+    """Show a corpus's stored config and stats."""
+    corpus = open_corpus(name)
+    cfg = corpus.store.get_config()
+    reg = registry.get(name)
+    return f"name: {name}\nregistered: {reg}\nrecords: {len(corpus)}\nconfig: {cfg}"
+
+
+def rm(name):
+    """Unregister a corpus (does not delete its built data)."""
+    registry.unregister(name)
+    return f"unregistered {name!r}"
+
+
+COMMANDS = [ls, register, build, search, info, rm]
diff --git a/ir/registry.py b/ir/registry.py
new file mode 100644
index 0000000..ca6bda6
--- /dev/null
+++ b/ir/registry.py
@@ -0,0 +1,99 @@
+"""Named-corpus registry — persistent, reusable corpus definitions.
+
+A registry entry records *how* to (re)build a corpus — its ``kind`` (a source
+preset), parameters, and embedder spec — so a corpus becomes a stable name you
+can build once and query across sessions. The registry is a single JSON file
+under the config dir (``~/.config/ir/corpora.json``).
+
+Presets map to :class:`~ir.sources.CorpusSource` constructors:
+
+- ``skills``   → :meth:`CorpusSource.from_skills`
+- ``packages`` → :meth:`CorpusSource.from_packages`
+- ``reports``  → :meth:`CorpusSource.from_md_reports`
+- ``files``    → :meth:`CorpusSource.from_files` (needs ``root``; optional
+  ``pattern``)
+
+Unregistered preset names (``skills``/``packages``/``reports``) are
+auto-registered with defaults on first use, so ``ir build skills`` just works.
+"""
+
+from __future__ import annotations
+
+import json
+from typing import Any
+
+from .config import registry_path
+from .sources import CorpusSource
+
+PRESETS = ("skills", "packages", "reports")
+
+
+def _load() -> dict[str, Any]:
+    path = registry_path()
+    if path.exists():
+        return json.loads(path.read_text(encoding="utf-8"))
+    return {}
+
+
+def _save(entries: dict[str, Any]) -> None:
+    registry_path().write_text(json.dumps(entries, indent=2), encoding="utf-8")
+
+
+def register(name: str, kind: str, *, embedder: str = "default", **params) -> dict:
+    """Register (or overwrite) a named corpus definition."""
+    if kind not in PRESETS and kind != "files":
+        raise ValueError(
+            f"Unknown corpus kind {kind!r}; use one of {PRESETS} or 'files'."
+        )
+    entries = _load()
+    entries[name] = {"kind": kind, "embedder": embedder, "params": params}
+    _save(entries)
+    return entries[name]
+
+
+def registered() -> dict[str, Any]:
+    """All registered corpus definitions, keyed by name."""
+    return _load()
+
+
+def get(name: str) -> dict | None:
+    """The registry entry for *name*, or ``None``."""
+    return _load().get(name)
+
+
+def unregister(name: str) -> None:
+    """Remove *name* from the registry (does not delete built data)."""
+    entries = _load()
+    entries.pop(name, None)
+    _save(entries)
+
+
+def source_from_entry(name: str, entry: dict) -> CorpusSource:
+    """Reconstruct a :class:`CorpusSource` from a registry entry."""
+    kind = entry["kind"]
+    params = dict(entry.get("params", {}))
+    embedder = entry.get("embedder", "default")
+    if kind == "skills":
+        return CorpusSource.from_skills(name=name, embedder=embedder)
+    if kind == "packages":
+        return CorpusSource.from_packages(name=name, embedder=embedder)
+    if kind == "reports":
+        return CorpusSource.from_md_reports(name=name, embedder=embedder)
+    if kind == "files":
+        root = params.pop("root")
+        return CorpusSource.from_files(root, name=name, embedder=embedder, **params)
+    raise ValueError(f"Unknown corpus kind {kind!r}.")
+
+
+def source_for(name: str) -> CorpusSource:
+    """Resolve *name* to a source, auto-registering a preset if needed."""
+    entry = get(name)
+    if entry is None:
+        if name in PRESETS:
+            entry = register(name, name)
+        else:
+            raise KeyError(
+                f"Corpus {name!r} is not registered. Register it with "
+                f"`ir register`, or use a preset name: {PRESETS}."
+            )
+    return source_from_entry(name, entry)
diff --git a/pyproject.toml b/pyproject.toml
index 6a3eb65..0f9bca0 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -6,7 +6,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "ir"
-version = "0.0.1"
+version = "0.1.0"
 description = "Information Retrieval"
 readme = "README.md"
 requires-python = ">=3.10"
@@ -17,8 +17,12 @@ dependencies = [
     "dol",
     "ef",
     "vd",
+    "argh",
 ]
 
+[project.scripts]
+ir = "ir.__main__:main"
+
 [project.license]
 text = "mit"
 
diff --git a/tests/test_registry.py b/tests/test_registry.py
new file mode 100644
index 0000000..c505d7d
--- /dev/null
+++ b/tests/test_registry.py
@@ -0,0 +1,72 @@
+"""Tests for the named-corpus registry, facade, and CLI (hermetic)."""
+
+import ir
+from ir import cli, registry
+
+
+def _isolate(tmp_path, monkeypatch):
+    monkeypatch.setenv("IR_CONFIG_DIR", str(tmp_path / "config"))
+    monkeypatch.setenv("IR_DATA_DIR", str(tmp_path / "data"))
+    monkeypatch.setenv("IR_CACHE_DIR", str(tmp_path / "cache"))
+
+
+def test_register_get_unregister(tmp_path, monkeypatch):
+    _isolate(tmp_path, monkeypatch)
+    registry.register("foo", "files", root="/some/dir", pattern=r".*\.md$")
+    assert registry.get("foo")["kind"] == "files"
+    assert "foo" in registry.registered()
+    registry.unregister("foo")
+    assert registry.get("foo") is None
+
+
+def test_register_rejects_unknown_kind(tmp_path, monkeypatch):
+    _isolate(tmp_path, monkeypatch)
+    try:
+        registry.register("bad", "nonsense")
+    except ValueError:
+        return
+    raise AssertionError("expected ValueError for unknown kind")
+
+
+def test_files_corpus_build_search_roundtrip(tmp_path, monkeypatch):
+    _isolate(tmp_path, monkeypatch)
+    docs = tmp_path / "docs"
+    docs.mkdir()
+    (docs / "deploy.md").write_text("How to deploy the app to the server with systemd.")
+    (docs / "baking.md").write_text(
+        "Recipe to bake a cake in the oven with flour sugar."
+    )
+
+    ir.register("notes", "files", root=str(docs), pattern=r".*\.md$")
+    corpus = ir.build_corpus("notes", embedder="light")
+    assert len(corpus) >= 2
+
+    hits = ir.search("notes", "bake a cake in the oven", k=1)
+    assert hits[0].artifact_id == "baking.md"
+
+
+def test_source_for_auto_registers_preset(tmp_path, monkeypatch):
+    _isolate(tmp_path, monkeypatch)
+    # A preset name resolves even if not explicitly registered (auto-registered).
+    # Use the registry layer; building 'skills'/'packages' needs priv/$PP, so we
+    # only assert the registration happens, not the build.
+    try:
+        registry.source_for("reports")
+    except Exception:
+        pass  # building the source may need $PP; registration is what we check
+    assert registry.get("reports") is not None
+    assert registry.get("reports")["kind"] == "reports"
+
+
+def test_cli_commands_return_strings(tmp_path, monkeypatch):
+    _isolate(tmp_path, monkeypatch)
+    docs = tmp_path / "docs"
+    docs.mkdir()
+    (docs / "a.md").write_text("alpha widgets and gadgets for testing the cli output.")
+    cli.register("notes", "files", root=str(docs), pattern=r".*\.md$")
+    cli.build("notes", embedder="light")
+    assert "notes" in cli.ls()
+    assert "alpha" in cli.search("notes", "widgets", k=3) or "a.md" in cli.search(
+        "notes", "widgets", k=3
+    )
+    assert "records:" in cli.info("notes")