Skip to content

Codebase intelligence for AI and humans: code health scores, auto-generated docs, git analytics, dead code detection, and architectural decisions via MCP.

fix(embedder): make edenai selectable everywhere it is documented (#1820)

Victor M. SMITH0364ff6323e5
diff --git a/packages/cli/src/repowise/cli/commands/init_cmd/command.py b/packages/cli/src/repowise/cli/commands/init_cmd/command.py
index 2bdc82dc5..b6d623eb9 100644
--- a/packages/cli/src/repowise/cli/commands/init_cmd/command.py
+++ b/packages/cli/src/repowise/cli/commands/init_cmd/command.py
@@ -411,7 +411,7 @@ def _run_generation_phase(
default=None,
help=(
"LLM provider name (anthropic, openai, openrouter, gemini, "
- "deepseek, kimi, ollama, litellm, codex_cli, opencode, mock)."
+ "deepseek, kimi, ollama, litellm, codex_cli, opencode, edenai, mock)."
),
)
@click.option("--model", default=None, help="Model identifier override.")
@@ -419,8 +419,11 @@ def _run_generation_phase(
"--embedder",
"embedder_name",
default=None,
- type=click.Choice(["gemini", "openai", "openrouter", "ollama", "mock"]),
- help="Embedder for RAG: gemini | openai | openrouter | ollama | mock (default: auto-detect).",
+ type=click.Choice(["gemini", "openai", "openrouter", "ollama", "edenai", "mock"]),
+ help=(
+ "Embedder for RAG: gemini | openai | openrouter | ollama | edenai | mock "
+ "(default: auto-detect)."
+ ),
)
@click.option("--skip-tests", is_flag=True, default=False, help="Skip test files.")
@click.option("--skip-infra", is_flag=True, default=False, help="Skip infrastructure files.")
diff --git a/packages/cli/src/repowise/cli/commands/reindex_cmd.py b/packages/cli/src/repowise/cli/commands/reindex_cmd.py
index 113a313e3..3a81ef134 100644
--- a/packages/cli/src/repowise/cli/commands/reindex_cmd.py
+++ b/packages/cli/src/repowise/cli/commands/reindex_cmd.py
@@ -19,7 +19,7 @@ from repowise.cli.ui import BRAND_STYLE, OWL_SPINNER
@click.argument("path", required=False, default=None)
@click.option(
"--embedder",
- type=click.Choice(["gemini", "openai", "openrouter", "ollama", "mock", "auto"]),
+ type=click.Choice(["gemini", "openai", "openrouter", "ollama", "edenai", "mock", "auto"]),
default="auto",
help="Embedder to use. 'auto' detects from env vars / config.",
)
diff --git a/packages/cli/src/repowise/cli/ui/mode_selection.py b/packages/cli/src/repowise/cli/ui/mode_selection.py
index 76d7acade..0f4db64db 100644
--- a/packages/cli/src/repowise/cli/ui/mode_selection.py
+++ b/packages/cli/src/repowise/cli/ui/mode_selection.py
@@ -450,7 +450,7 @@ def _prompt_generation(
# Embedder selection
detected_embedder = _resolve_embedder_from_env()
- embedder_choices = ["gemini", "openai", "openrouter", "ollama", "mock"]
+ embedder_choices = ["gemini", "openai", "openrouter", "ollama", "edenai", "mock"]
result["embedder"] = click.prompt(
" Embedder for RAG",
default=detected_embedder,
@@ -644,7 +644,7 @@ def _prompt_index_only_search(console: Console, result: dict[str, Any]) -> None:
result["embedder"] = click.prompt(
" Embedder for semantic search (mock = full-text only)",
default="mock",
- type=click.Choice(["mock", "ollama", "gemini", "openai", "openrouter"]),
+ type=click.Choice(["mock", "ollama", "gemini", "openai", "openrouter", "edenai"]),
)
@@ -658,6 +658,11 @@ def _resolve_embedder_from_env() -> str:
return "openrouter"
if os.environ.get("OLLAMA_EMBEDDING_MODEL"):
return "ollama"
+ # Last, like the shared resolver in cli/providers/embedders.py: an unrelated
+ # EDENAI_API_KEY in the environment must not outrank a provider the user was
+ # already resolving to.
+ if os.environ.get("EDENAI_API_KEY"):
+ return "edenai"
return "mock"
diff --git a/packages/server/src/repowise/server/app.py b/packages/server/src/repowise/server/app.py
index ff454ce83..e01f8ca92 100644
--- a/packages/server/src/repowise/server/app.py
+++ b/packages/server/src/repowise/server/app.py
@@ -103,6 +103,7 @@ def _build_embedder():
gemini — GeminiEmbedder via GEMINI_API_KEY / GOOGLE_API_KEY env var
openai — OpenAIEmbedder via OPENAI_API_KEY env var
openrouter — OpenRouterEmbedder via OPENROUTER_API_KEY env var
+ edenai — EdenAIEmbedder via EDENAI_API_KEY env var
"""
name = os.environ.get("REPOWISE_EMBEDDER", "mock").lower()
if name == "ollama":
@@ -129,8 +130,14 @@ def _build_embedder():
model = os.environ.get("REPOWISE_EMBEDDING_MODEL", "google/gemini-embedding-001")
return OpenRouterEmbedder(model=model)
+ if name == "edenai":
+ from repowise.core.providers.embedding.edenai import EdenAIEmbedder
+
+ model = os.environ.get("REPOWISE_EMBEDDING_MODEL", "amazon/amazon.titan-embed-text-v2:0")
+ return EdenAIEmbedder(model=model)
logger.warning(
- "embedder.mock_active: set REPOWISE_EMBEDDER=gemini, openai, openrouter, or ollama for real RAG"
+ "embedder.mock_active: set REPOWISE_EMBEDDER=gemini, openai, openrouter, "
+ "ollama, or edenai for real RAG"
)
return KeylessEmbedder()
diff --git a/tests/unit/cli/test_edenai_reachability.py b/tests/unit/cli/test_edenai_reachability.py
new file mode 100644
index 000000000..5547bfaf4
--- /dev/null
+++ b/tests/unit/cli/test_edenai_reachability.py
@@ -0,0 +1,133 @@
+"""The Eden AI embedder has to be reachable from `repowise init`, not just registered.
+
+``docs/reference/CLI_REFERENCE.md`` documents ``--embedder edenai``, and the
+embedder itself is in the registry, but the flag's ``click.Choice`` and both
+interactive pickers were left with the older five-backend list. The flag was
+rejected outright, the pickers never offered the option, and a machine whose
+only key is ``EDENAI_API_KEY`` defaulted to the mock, so semantic search ran on
+vectors that cannot match the index.
+
+Precedence matters as much as presence: ``EDENAI_API_KEY`` is read last, so a
+key sitting in the environment for something else cannot outrank a backend the
+user was already resolving to.
+"""
+
+from __future__ import annotations
+
+import click
+import pytest
+
+from repowise.cli.commands.init_cmd.command import init_command
+from repowise.cli.commands.reindex_cmd import reindex_command
+from repowise.cli.ui import mode_selection
+
+_DETECTION_KEYS = (
+ "GEMINI_API_KEY",
+ "GOOGLE_API_KEY",
+ "OPENAI_API_KEY",
+ "OPENROUTER_API_KEY",
+ "OLLAMA_EMBEDDING_MODEL",
+ "EDENAI_API_KEY",
+)
+
+
+def _clear_detection_env(monkeypatch: pytest.MonkeyPatch) -> None:
+ for key in _DETECTION_KEYS:
+ monkeypatch.delenv(key, raising=False)
+
+
+class _NullConsole:
+ """Same stand-in the existing UI tests use: the prompts are what matter."""
+
+ def print(self, *_args: object, **_kwargs: object) -> None:
+ pass
+
+
+class TestEmbedderFlag:
+ def test_init_accepts_edenai(self) -> None:
+ ctx = init_command.make_context("init", ["--embedder", "edenai"])
+ assert ctx.params["embedder_name"] == "edenai"
+
+ def test_reindex_accepts_edenai(self) -> None:
+ """`reindex` documents the same set as `init`, so it has to accept it too."""
+ ctx = reindex_command.make_context("reindex", ["--embedder", "edenai"])
+ assert ctx.params["embedder"] == "edenai"
+
+ @pytest.mark.parametrize("command", [init_command, reindex_command])
+ def test_an_unknown_backend_is_still_rejected(self, command: click.Command) -> None:
+ with pytest.raises(click.UsageError):
+ command.make_context(command.name or "cmd", ["--embedder", "not-a-backend"])
+
+
+class TestInteractivePickers:
+ """Whichever prompt asks for an embedder must list every real backend.
+
+ The two prompts are written separately, one in advanced config and one in
+ the index-only Search section, which is how they drifted apart.
+ """
+
+ @staticmethod
+ def _offered_choices(monkeypatch: pytest.MonkeyPatch, run) -> list[str]:
+ offered: list[str] = []
+
+ def _capture(_text, *_a, **kwargs):
+ choice_type = kwargs.get("type")
+ if isinstance(choice_type, click.Choice):
+ offered.extend(choice_type.choices)
+ # Answer every prompt with its own default, like the existing UI
+ # tests do: returning a coerced value breaks the numeric prompts.
+ return kwargs.get("default", "")
+
+ monkeypatch.setattr(mode_selection.click, "prompt", _capture)
+ monkeypatch.setattr(
+ mode_selection.click, "confirm", lambda *_a, **k: k.get("default", False)
+ )
+ _clear_detection_env(monkeypatch)
+ run()
+ return offered
+
+ def test_advanced_config_offers_edenai(self, monkeypatch: pytest.MonkeyPatch) -> None:
+ offered = self._offered_choices(
+ monkeypatch,
+ lambda: mode_selection.interactive_advanced_config(_NullConsole()),
+ )
+ assert "edenai" in offered
+
+ def test_index_only_search_offers_edenai(self, monkeypatch: pytest.MonkeyPatch) -> None:
+ offered = self._offered_choices(
+ monkeypatch,
+ lambda: mode_selection._prompt_index_only_search(_NullConsole(), {}),
+ )
+ assert "edenai" in offered
+
+
+class TestEnvDetection:
+ def test_the_key_alone_resolves_to_edenai(self, monkeypatch: pytest.MonkeyPatch) -> None:
+ _clear_detection_env(monkeypatch)
+ monkeypatch.setenv("EDENAI_API_KEY", "x")
+ assert mode_selection._resolve_embedder_from_env() == "edenai"
+
+ def test_no_key_still_resolves_to_mock(self, monkeypatch: pytest.MonkeyPatch) -> None:
+ _clear_detection_env(monkeypatch)
+ assert mode_selection._resolve_embedder_from_env() == "mock"
+
+ @pytest.mark.parametrize(
+ ("other_key", "other_value", "expected"),
+ [
+ ("GEMINI_API_KEY", "x", "gemini"),
+ ("OPENAI_API_KEY", "x", "openai"),
+ ("OPENROUTER_API_KEY", "x", "openrouter"),
+ ("OLLAMA_EMBEDDING_MODEL", "embeddinggemma", "ollama"),
+ ],
+ )
+ def test_edenai_never_outranks_an_incumbent(
+ self,
+ monkeypatch: pytest.MonkeyPatch,
+ other_key: str,
+ other_value: str,
+ expected: str,
+ ) -> None:
+ _clear_detection_env(monkeypatch)
+ monkeypatch.setenv("EDENAI_API_KEY", "x")
+ monkeypatch.setenv(other_key, other_value)
+ assert mode_selection._resolve_embedder_from_env() == expected
diff --git a/tests/unit/cli/test_shared_helpers.py b/tests/unit/cli/test_shared_helpers.py
index dfad2be55..ade2b5042 100644
--- a/tests/unit/cli/test_shared_helpers.py
+++ b/tests/unit/cli/test_shared_helpers.py
@@ -47,6 +47,8 @@ def test_resolve_embedder_explicit_flag_wins(monkeypatch: pytest.MonkeyPatch) ->
({"OPENROUTER_API_KEY": "x"}, "openrouter"),
({"OLLAMA_BASE_URL": "http://localhost:11434"}, "mock"),
({"OLLAMA_EMBEDDING_MODEL": "embeddinggemma"}, "ollama"),
+ ({"EDENAI_API_KEY": "x"}, "edenai"),
+ ({"OLLAMA_EMBEDDING_MODEL": "embeddinggemma", "EDENAI_API_KEY": "x"}, "ollama"),
({"REPOWISE_EMBEDDER": "ollama"}, "ollama"),
({}, "mock"),
],
@@ -61,6 +63,7 @@ def test_resolve_embedder_env_detection(
"OPENROUTER_API_KEY",
"OLLAMA_BASE_URL",
"OLLAMA_EMBEDDING_MODEL",
+ "EDENAI_API_KEY",
"REPOWISE_EMBEDDER",
):
monkeypatch.delenv(key, raising=False)
diff --git a/tests/unit/server/test_app_embedder_backends.py b/tests/unit/server/test_app_embedder_backends.py
new file mode 100644
index 000000000..89091a10c
--- /dev/null
+++ b/tests/unit/server/test_app_embedder_backends.py
@@ -0,0 +1,48 @@
+"""`_build_embedder` must honour every documented `REPOWISE_EMBEDDER` value.
+
+``docs/reference/CONFIG.md`` lists ``edenai`` among the accepted values, but the
+HTTP server resolved backends through a hardcoded chain that stopped at
+``openrouter``, so ``REPOWISE_EMBEDDER=edenai`` fell through to the keyless
+mock with only a warning: the server then reported healthy while semantic
+search ran on vectors that cannot match the index. That is the same failure the
+MCP server fixed in #324 by going through the shared registry; this chain is
+the remaining hardcoded copy.
+"""
+
+from __future__ import annotations
+
+import pytest
+
+from repowise.core.providers.embedding.base import KeylessEmbedder
+from repowise.core.providers.embedding.edenai import EdenAIEmbedder
+from repowise.server.app import _build_embedder
+
+
[email protected](autouse=True)
+def _clean_embedder_env(monkeypatch: pytest.MonkeyPatch) -> None:
+ for key in ("REPOWISE_EMBEDDER", "REPOWISE_EMBEDDING_MODEL", "EDENAI_API_KEY"):
+ monkeypatch.delenv(key, raising=False)
+
+
+def test_edenai_builds_a_real_embedder(monkeypatch: pytest.MonkeyPatch) -> None:
+ monkeypatch.setenv("REPOWISE_EMBEDDER", "edenai")
+ monkeypatch.setenv("EDENAI_API_KEY", "test-key-not-a-real-one")
+ assert isinstance(_build_embedder(), EdenAIEmbedder)
+
+
+def test_edenai_honours_the_indexed_model(monkeypatch: pytest.MonkeyPatch) -> None:
+ """Same guard as the other backends: serve must not silently re-embed at a
+ different width than the one init wrote (issue #426)."""
+ monkeypatch.setenv("REPOWISE_EMBEDDER", "edenai")
+ monkeypatch.setenv("EDENAI_API_KEY", "test-key-not-a-real-one")
+ monkeypatch.setenv("REPOWISE_EMBEDDING_MODEL", "openai/text-embedding-3-large")
+ embedder = _build_embedder()
+ assert isinstance(embedder, EdenAIEmbedder)
+ assert embedder.dimensions == 3072
+
+
+def test_an_unknown_backend_still_degrades_to_the_keyless_mock(
+ monkeypatch: pytest.MonkeyPatch,
+) -> None:
+ monkeypatch.setenv("REPOWISE_EMBEDDER", "not-a-backend")
+ assert isinstance(_build_embedder(), KeylessEmbedder)