Skip to content

Hypergraph is more powerful. Transform unstructured text into structured knowledge with LLMs. Graphs, hypergraphs, and spatio-temporal extractions — with one command.

fix: do not write a same-provider embedder on LLM-only config init

Frankie-Xu85307c3cac7e
diff --git a/hyperextract/cli/commands/config.py b/hyperextract/cli/commands/config.py
index 07cd71c..c10e4b9 100644
--- a/hyperextract/cli/commands/config.py
+++ b/hyperextract/cli/commands/config.py
@@ -12,6 +12,15 @@ from ..config import ConfigManager
logger = get_logger("he.config")
console = Console()
+
+def _preset_embedder_model(preset: dict) -> str | None:
+ """Return a usable default embedder model, or None for LLM-only presets."""
+ model = preset.get("default_embedder")
+ if isinstance(model, str) and model.strip():
+ return model
+ return None
+
+
app = typer.Typer(
name="config",
help="Manage LLM and Embedder configuration",
@@ -304,7 +313,7 @@ def init(
preset = PROVIDER_PRESETS.get(provider, {})
llm_model = preset.get("default_llm") or "gpt-4o-mini"
- emb_model = preset.get("default_embedder") or "text-embedding-3-small"
+ emb_model = _preset_embedder_model(preset)
preset_url = preset.get("base_url") or ""
resolved_base = base_url or preset_url
@@ -323,7 +332,9 @@ def init(
)
else:
console.print(
- f"[yellow]Warning: Provider '{provider}' has no default embedder. Please configure embedder separately.[/yellow]"
+ f"[yellow]Warning: Provider '{provider}' has no default embedder. "
+ "Please configure embedder separately: "
+ "he config embedder -p openai -k … (or vLLM).[/yellow]"
)
console.print("[bold green]Configuration saved successfully![/bold green]")
@@ -390,7 +401,7 @@ def init(
preset = PROVIDER_PRESETS.get(selected, {})
preset_url = preset.get("base_url") or ""
default_llm = preset.get("default_llm") or "gpt-4o-mini"
- default_emb = preset.get("default_embedder") or "text-embedding-3-small"
+ default_emb = _preset_embedder_model(preset)
console.print()
console.print(f"[bold]Step 2: LLM Configuration (Provider: {selected})[/bold]")
@@ -435,11 +446,37 @@ def init(
console.print("[bold]Step 3: Embedder Configuration[/bold]")
if selected == "vllm":
+ emb_provider = selected
emb_model = console.input(" Embedder Model (e.g. bge-m3): ").strip()
emb_base_url = console.input(
" Embedder Base URL (e.g. http://localhost:8001/v1): "
).strip()
+ elif default_emb is None:
+ console.print(
+ f" [yellow]Provider '{selected}' has no default embedder. "
+ "Configure an OpenAI-compatible embedder separately "
+ "(default provider: openai).[/yellow]"
+ )
+ emb_provider = (
+ console.input(" Embedder provider (default: openai): ").strip() or "openai"
+ )
+ emb_preset = PROVIDER_PRESETS.get(emb_provider, {})
+ emb_default_model = (
+ _preset_embedder_model(emb_preset) or "text-embedding-3-small"
+ )
+ emb_preset_url = emb_preset.get("base_url") or ""
+ emb_model = (
+ console.input(f" Model (default: {emb_default_model}): ").strip()
+ or emb_default_model
+ )
+ emb_base_url = (
+ console.input(
+ f" Base URL (default: {emb_preset_url}, press Enter to skip): "
+ ).strip()
+ or emb_preset_url
+ )
else:
+ emb_provider = selected
emb_model = (
console.input(f" Model (default: {default_emb}): ").strip() or default_emb
)
@@ -463,7 +500,7 @@ def init(
)
config.set_embedder(
- provider=selected,
+ provider=emb_provider,
model=emb_model,
api_key=emb_api_key,
base_url=emb_base_url or None,
diff --git a/tests/cli/test_config.py b/tests/cli/test_config.py
index 9a90498..75124e7 100644
--- a/tests/cli/test_config.py
+++ b/tests/cli/test_config.py
@@ -137,6 +137,8 @@ def test_load_warns_when_everyone_can_read(tmp_path, caplog):
assert caplog.records
assert any("Everyone" in record.getMessage() for record in caplog.records)
assert _FAKE_API_KEY not in caplog.text
+
+
def test_cli_provider_tables_are_library_presets():
"""CLI must not keep a drifting copy of library provider tables."""
from hyperextract.cli import config as cli_config
@@ -205,3 +207,103 @@ def test_get_llm_config_reads_google_env_key(tmp_path, monkeypatch):
assert cfg.api_key == "sk-google-from-env"
assert cfg.base_url == ""
+
+
+def _isolate_default_config(tmp_path, monkeypatch):
+ """Point ConfigManager() at a temp toml so CLI init never touches ~/.he."""
+ cfg_path = tmp_path / "config.toml"
+ monkeypatch.setattr("hyperextract.cli.config.DEFAULT_CONFIG_FILE", cfg_path)
+ return cfg_path
+
+
+def _read_init_toml(cfg_path):
+ import tomllib
+
+ with open(cfg_path, "rb") as f:
+ return tomllib.load(f)
+
+
[email protected]("provider", ["google", "anthropic", "deepseek"])
+def test_quick_init_llm_only_provider_does_not_write_same_provider_embedder(
+ tmp_path, monkeypatch, provider
+):
+ """Quick init must not copy an LLM-only provider onto embedder.provider."""
+ from typer.testing import CliRunner
+
+ from hyperextract.cli.cli import app
+ from hyperextract.utils.client import PROVIDER_PRESETS
+
+ cfg_path = _isolate_default_config(tmp_path, monkeypatch)
+ runner = CliRunner()
+ result = runner.invoke(app, ["config", "init", "-p", provider, "-k", _FAKE_API_KEY])
+
+ assert result.exit_code == 0, result.output
+ assert "no default embedder" in result.output
+ assert "he config embedder" in result.output
+
+ data = _read_init_toml(cfg_path)
+ assert data["llm"]["provider"] == provider
+ assert data["llm"]["model"] == PROVIDER_PRESETS[provider]["default_llm"]
+ assert data["llm"]["api_key"] == _FAKE_API_KEY
+ assert data["embedder"].get("provider") != provider
+ assert data["embedder"].get("api_key") in ("", None)
+
+
+def test_quick_init_openai_still_writes_same_provider_embedder(tmp_path, monkeypatch):
+ """Providers with a real default_embedder still share the LLM provider."""
+ from typer.testing import CliRunner
+
+ from hyperextract.cli.cli import app
+
+ cfg_path = _isolate_default_config(tmp_path, monkeypatch)
+ runner = CliRunner()
+ result = runner.invoke(app, ["config", "init", "-p", "openai", "-k", _FAKE_API_KEY])
+
+ assert result.exit_code == 0, result.output
+ data = _read_init_toml(cfg_path)
+ assert data["llm"]["provider"] == "openai"
+ assert data["embedder"]["provider"] == "openai"
+ assert data["embedder"]["model"] == "text-embedding-3-small"
+ assert data["embedder"]["api_key"] == _FAKE_API_KEY
+
+
+def test_interactive_init_llm_only_prompts_separate_openai_embedder(
+ tmp_path, monkeypatch
+):
+ """Interactive Google init must not keep embedder.provider == google."""
+ from hyperextract.cli.commands import config as config_cmd
+
+ cfg_path = _isolate_default_config(tmp_path, monkeypatch)
+ answers = iter(
+ [
+ "6", # google in the interactive provider list
+ "", # default LLM model
+ "", # default LLM base URL
+ _FAKE_API_KEY,
+ "", # default embedder provider (openai)
+ "", # default embedder model
+ "", # default embedder base URL
+ "sk-emb",
+ ]
+ )
+ monkeypatch.setattr(config_cmd.console, "input", lambda _prompt="": next(answers))
+
+ config_cmd.init(provider=None, api_key=None, base_url=None)
+
+ data = _read_init_toml(cfg_path)
+ assert data["llm"]["provider"] == "google"
+ assert data["llm"]["model"] == "gemini-3.8-flash"
+ assert data["embedder"]["provider"] == "openai"
+ assert data["embedder"]["model"] == "text-embedding-3-small"
+ assert data["embedder"]["api_key"] == "sk-emb"
+
+
+def test_legacy_broken_google_embedder_still_fails_create_embedder():
+ """Lock the create_embedder guard so a leftover google embedder still fails."""
+ from hyperextract.utils.client import create_embedder
+
+ with pytest.raises(ValueError, match="embed"):
+ create_embedder(
+ {"provider": "google", "model": "text-embedding-3-small"},
+ api_key=_FAKE_API_KEY,
+ )