fix: respect OPENAI_BASE_URL for openai embedding provider

[paralizeer]

Problem

The openai embedding provider hardcodes the base URL to OpenAI's default endpoint:

self.client = AsyncOpenAI(api_key=api_key)  # always hits api.openai.com

This makes it impossible to use OpenAI-compatible embedding servers like Ollama, local inference servers, or self-hosted models.

Fix

Add two new fields to LLMSettings:

• OPENAI_BASE_URL: passed as base_url to AsyncOpenAI, enabling any OpenAI-compatible endpoint
• OPENAI_EMBEDDING_MODEL: override the model name (e.g. nomic-embed-text for Ollama)

Example usage

# config.toml — use Ollama for embeddings
[llm]
OPENAI_API_KEY = "ollama"
OPENAI_BASE_URL = "http://localhost:11434/v1"
OPENAI_EMBEDDING_MODEL = "nomic-embed-text"  # 768-dim model

Testing

Tested with Ollama + nomic-embed-text (768 dims) producing correct embeddings stored in pgvector.

Summary by CodeRabbit

• New Features
  • Added optional configuration for a custom OpenAI API base URL to support alternate endpoints.
  • Added optional configuration to choose the OpenAI embedding model, with a sensible default used when not specified.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: be446fe7-6e2e-4a4e-8277-6bf3469df8dc

📥 Commits

Reviewing files that changed from the base of the PR and between b75e0ea and 02f9f45.

📒 Files selected for processing (1)

• src/embedding_client.py

🚧 Files skipped from review as they are similar to previous changes (1)

• src/embedding_client.py

---

Walkthrough

Adds two optional LLM settings—OPENAI_BASE_URL and OPENAI_EMBEDDING_MODEL—and updates the OpenAI embedding client to use these settings for client initialization and default embedding model selection.

Changes

Cohort / File(s)	Summary
Configuration Settings src/config.py	Added optional fields OPENAI_BASE_URL and OPENAI_EMBEDDING_MODEL to LLMSettings.
Embedding Client Initialization src/embedding_client.py	OpenAI branch now reads settings.LLM.OPENAI_BASE_URL and passes it as base_url to AsyncOpenAI; selects embedding model from settings.LLM.OPENAI_EMBEDDING_MODEL with fallback to "text-embedding-3-small".

Estimated Code Review Effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Poem

🐰 I nibble configs with a twitch and twirl,
New URLs and models set my whiskers in a whirl,
Embeddings choose their path, neat and small,
I hop away delighted—happy changes for all! ✨

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately summarizes the main change: adding support for custom OPENAI_BASE_URL in the OpenAI embedding provider, which aligns with the core problem and fix described in the PR objectives.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

🧹 Nitpick comments (1)

src/embedding_client.py (1)

67-67: Prefer direct attribute access over getattr.

Since OPENAI_EMBEDDING_MODEL is now a defined field in LLMSettings with a default of None, you can use direct attribute access for consistency with line 63.

Suggested change

-            self.model = getattr(settings.LLM, "OPENAI_EMBEDDING_MODEL", None) or "text-embedding-3-small"
+            self.model = settings.LLM.OPENAI_EMBEDDING_MODEL or "text-embedding-3-small"

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@src/embedding_client.py` at line 67, Replace the getattr usage with direct
attribute access: set self.model = settings.LLM.OPENAI_EMBEDDING_MODEL or
"text-embedding-3-small" so the code uses the defined LLMSettings field
(consistent with the access pattern on line 63); update the assignment where
self.model is initialized (in the Embedding client constructor or the method
that sets self.model) to use settings.LLM.OPENAI_EMBEDDING_MODEL instead of
getattr(settings.LLM, "OPENAI_EMBEDDING_MODEL", None).

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Nitpick comments:
In `@src/embedding_client.py`:
- Line 67: Replace the getattr usage with direct attribute access: set
self.model = settings.LLM.OPENAI_EMBEDDING_MODEL or "text-embedding-3-small" so
the code uses the defined LLMSettings field (consistent with the access pattern
on line 63); update the assignment where self.model is initialized (in the
Embedding client constructor or the method that sets self.model) to use
settings.LLM.OPENAI_EMBEDDING_MODEL instead of getattr(settings.LLM,
"OPENAI_EMBEDDING_MODEL", None).

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 0071fb87-3145-466b-933a-30545274468d

📥 Commits

Reviewing files that changed from the base of the PR and between a5423b5 and b75e0ea.

📒 Files selected for processing (2)

• src/config.py
• src/embedding_client.py

[paralizeer]

Applied the suggestion from @coderabbitai — replaced getattr with direct attribute access since OPENAI_EMBEDDING_MODEL is already a defined field in LLMSettings. Cleaner and consistent with the rest of the file.

Context: this fix came out of integrating Honcho with an Ollama-compatible embedding endpoint (nomic-embed-text). The OPENAI_BASE_URL override is what makes it work with non-OpenAI providers.

[coderabbitai]

Tip

For best results, initiate chat on the files or code changes.

@paralizeer: Thanks for applying the change and for the helpful context! Direct attribute access is indeed cleaner here since OPENAI_EMBEDDING_MODEL is a first-class field on LLMSettings. The Ollama + nomic-embed-text use-case is a great motivating example for this fix — good to have OPENAI_BASE_URL as the primary lever for pointing at any OpenAI-compatible endpoint.

---

✏️ Learnings added

Learnt from: paralizeer
URL: https://github.com/plastic-labs/honcho/pull/464

Timestamp: 2026-03-30T17:35:06.547Z
Learning: In the Honcho codebase (src/embedding_client.py and src/config.py), `OPENAI_BASE_URL` and `OPENAI_EMBEDDING_MODEL` are first-class fields defined on `LLMSettings`. Access them via direct attribute access (e.g., `settings.LLM.OPENAI_BASE_URL`) rather than `getattr`. `OPENAI_BASE_URL` is the key override for using OpenAI-compatible embedding providers such as Ollama (e.g., base_url="http://localhost:11434/v1"), and `OPENAI_EMBEDDING_MODEL` allows specifying a custom model name (e.g., "nomic-embed-text") instead of the default "text-embedding-3-small".

Note: Learnings are effective only in the context of similar code segments. To apply general coding guidelines, please configure review instructions. You can manage existing learnings in the app.