fix(embedding): derive OpenAI provider dimensions from model

[rohitg00]

Summary

CodeRabbit flagged this on #186: the PR made OPENAI_EMBEDDING_MODEL configurable but left dimensions hardcoded at 1536. A user pointing the embedding provider at text-embedding-3-large (3072 dims) would still see provider.dimensions === 1536, silently breaking downstream vector-store schemas and hybrid-search weighting.

Fix

• Add a MODEL_DIMENSIONS lookup for known OpenAI models:
  • text-embedding-3-small → 1536
  • text-embedding-3-large → 3072
  • text-embedding-ada-002 → 1536
• dimensions becomes a computed readonly field set from the lookup in the constructor.
• New OPENAI_EMBEDDING_DIMENSIONS env var explicitly overrides for custom / self-hosted OpenAI-compatible endpoints. Positive integers only; not-a-number, -5, 0 throw a clear error at construction.
• Unknown model names fall back to 1536 (the old default) so existing deployments are unchanged.

Tests

827 / 827 pass (+4 new):

• dimensions derived from 3 known models
• override via env var
• unknown-model fallback
• invalid override values rejected

Notes

Non-breaking. Preserves #186's defaults for users who did not set any embedding env vars.

Summary by CodeRabbit

• New Features
  • Added OPENAI_EMBEDDING_DIMENSIONS environment variable to customize embedding vector dimensions for OpenAI embeddings.
  • Embedding dimensions are now automatically determined based on your selected OpenAI embedding model, with built-in support for standard model types.
  • Enhanced configuration validation to detect invalid dimension settings with informative error messages, ensuring proper setup.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
agentmemory	Ready	Preview, Comment	Apr 22, 2026 11:39am

[coderabbitai]

📝 Walkthrough

Walkthrough

The OpenAI embedding provider now dynamically resolves embedding vector dimensions from environment configuration or model defaults. A lookup table and validation helper determine dimensions based on the model and an optional OPENAI_EMBEDDING_DIMENSIONS environment variable, which must be a positive integer when provided.

Changes

Cohort / File(s)	Summary
Embedding Provider Implementation src/providers/embedding/openai.ts	Added MODEL_DIMENSIONS lookup table, resolveDimensions() helper function, and runtime validation for OPENAI_EMBEDDING_DIMENSIONS environment variable. Changed dimensions from a fixed constant to dynamically initialized value with fallback to 1536.
Embedding Provider Tests test/embedding-provider.test.ts	Extended test setup to clear OPENAI_EMBEDDING_DIMENSIONS before each test. Added comprehensive unit tests for dimension selection across known models, override behavior, unknown model fallback, and validation error cases.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Poem

🐰 Dimensions dance in a rabbit's way,
Embeddings configured day by day,
With lookup tables crisp and clean,
The finest vectors ever seen!
Environment whispers 1536 or more,
Validation guards the provider's door. ✨

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 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: deriving OpenAI embedding provider dimensions from the configured model instead of using a hardcoded value.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

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

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/openai-embedding-dimensions

---

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]

Actionable comments posted: 1

🧹 Nitpick comments (1)

test/embedding-provider.test.ts (1)

108-150: Good coverage of the new behavior.

The four new cases (known-model derivation, override, unknown-model fallback, invalid-override rejection) map cleanly onto the new logic. One gap worth considering: add a case for a partial-numeric override like "768abc" or "1.5" to lock down stricter integer validation (see the comment on resolveDimensions).

🤖 Prompt for AI Agents

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

In `@test/embedding-provider.test.ts` around lines 108 - 150, Add tests for
partial-numeric and decimal OPENAI_EMBEDDING_DIMENSIONS to ensure
resolveDimensions rejects them: in embedding-provider.test.ts extend the
"rejects invalid OPENAI_EMBEDDING_DIMENSIONS values" case to set
process.env["OPENAI_EMBEDDING_DIMENSIONS"] = "768abc" and = "1.5" (and assert
new OpenAIEmbeddingProvider("test-key") throws the same
/OPENAI_EMBEDDING_DIMENSIONS must be a positive integer/), referencing the
OpenAIEmbeddingProvider constructor and resolveDimensions logic to validate
stricter integer parsing.

🤖 Prompt for all review comments with AI agents

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

Inline comments:
In `@src/providers/embedding/openai.ts`:
- Around line 20-31: The resolveDimensions function currently uses parseInt
which accepts partial/fractional inputs; change the override parsing to reject
any non-integer or extra-character strings by converting the trimmed override
with Number() and validating with Number.isInteger(parsed) && parsed > 0 (or use
a strict /^\d+$/ regex before Number conversion) instead of parseInt, and throw
the same error message when validation fails so fractional values like "1.5" or
"768abc" are rejected.

---

Nitpick comments:
In `@test/embedding-provider.test.ts`:
- Around line 108-150: Add tests for partial-numeric and decimal
OPENAI_EMBEDDING_DIMENSIONS to ensure resolveDimensions rejects them: in
embedding-provider.test.ts extend the "rejects invalid
OPENAI_EMBEDDING_DIMENSIONS values" case to set
process.env["OPENAI_EMBEDDING_DIMENSIONS"] = "768abc" and = "1.5" (and assert
new OpenAIEmbeddingProvider("test-key") throws the same
/OPENAI_EMBEDDING_DIMENSIONS must be a positive integer/), referencing the
OpenAIEmbeddingProvider constructor and resolveDimensions logic to validate
stricter integer parsing.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 9acee8a3-4b72-4363-882c-cc4d02e76848

📥 Commits

Reviewing files that changed from the base of the PR and between 343b51c and a6ce50d.

📒 Files selected for processing (2)

• src/providers/embedding/openai.ts
• test/embedding-provider.test.ts

[coderabbitai]

⚠️ Potential issue | 🟡 Minor

parseInt silently accepts partial-numeric and fractional inputs.

parseInt("768abc", 10) returns 768 and parseInt("1.5", 10) returns 1 — both bypass the validation and produce a silently-wrong dimensions value that will later cause dimension mismatches against the vector store. The current tests ("not-a-number", "-5", "0") don't catch this because fully non-numeric strings yield NaN.

Consider using Number() (or a regex pre-check) so any non-integer string is rejected.

🔧 Proposed fix

 function resolveDimensions(model: string, override: string | undefined): number {
-  if (override !== undefined && override.trim().length > 0) {
-    const parsed = parseInt(override, 10);
-    if (!Number.isFinite(parsed) || parsed <= 0) {
+  if (override !== undefined && override.trim().length > 0) {
+    const parsed = Number(override);
+    if (!Number.isInteger(parsed) || parsed <= 0) {
       throw new Error(
         `OPENAI_EMBEDDING_DIMENSIONS must be a positive integer, got: ${override}`,
       );
     }
     return parsed;
   }
   return MODEL_DIMENSIONS[model] ?? DEFAULT_DIMENSIONS;
 }

🤖 Prompt for AI Agents

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

In `@src/providers/embedding/openai.ts` around lines 20 - 31, The
resolveDimensions function currently uses parseInt which accepts
partial/fractional inputs; change the override parsing to reject any non-integer
or extra-character strings by converting the trimmed override with Number() and
validating with Number.isInteger(parsed) && parsed > 0 (or use a strict /^\d+$/
regex before Number conversion) instead of parseInt, and throw the same error
message when validation fails so fractional values like "1.5" or "768abc" are
rejected.