feat(embeddings): native Ollama embedding provider by PatrickSys · Pull Request #146 · PatrickSys/codebase-context

PatrickSys · 2026-10-30T22:27:48Z

Implements #71: EMBEDDING_PROVIDER=ollama uses a new OllamaEmbeddingProvider (/api/embed, default OLLAMA_HOST=http://localhost:11434, default model nomic-embed-text) instead of falling back to Transformers.js.

Skips @huggingface/transformers when provider is ollama; dimension hints for LanceDB via known model map + optional OLLAMA_EMBEDDING_DIMENSIONS
Config helpers in types.ts include OPENAI_BASE_URL wiring (compatible with First-run fixes: OPENAI_BASE_URL and CLI entrypoint regression test #145 / cursor/first-run-fixes-16ec) without duplicating that PR’s CLI entrypoint work
README Privacy section + docs/capabilities.md; Vitest HTTP mocks in tests/ollama-embeddings.test.ts and tests/embedding-config-env.test.ts

Fixes #71

Implement OllamaEmbeddingProvider using /api/embed with OLLAMA_HOST and provider-specific model defaults. Wire getEmbeddingProvider, dimension detection, and skip Transformers runtime for ollama. Includes OPENAI_BASE_URL config compatible with PR #145, docs, and HTTP-mocked tests.

chatgpt-codex-connector · 2026-10-30T22:28:01Z

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-30T22:29:03.947336Z	`333f520`	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 333f5203da

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-10-30T22:29:00Z

-    console.warn('Ollama provider not yet implemented, falling back to Transformers.js');
+    const { OllamaEmbeddingProvider } = await import('./ollama.js');
+    const provider = new OllamaEmbeddingProvider(
+      mergedConfig.model || DEFAULT_OLLAMA_MODEL,


Resolve the default model from the effective provider

When Ollama is selected through project configuration or getEmbeddingProvider({ provider: 'ollama' }), without setting EMBEDDING_PROVIDER, the merged configuration still contains the nonempty Transformers default Xenova/bge-small-en-v1.5. Consequently, this fallback never selects nomic-embed-text, and initialization requests the wrong model from Ollama, failing even when the documented default model is installed. Resolve provider-specific defaults after applying the explicit provider configuration; the indexer's configuration merge needs the same treatment.

Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-10-30T22:29:00Z

+      return parsed;
+    }
+  }
+  return 768;


Use discovered dimensions for incremental validation

For an Ollama model outside the hardcoded map that returns vectors other than 768 dimensions, a full build succeeds because embedBatch() discovers the actual vector length. However, the next incremental build calls getConfiguredDimensions(), which independently returns this 768 fallback unless OLLAMA_EMBEDDING_DIMENSIONS is set. LanceDBStorageProvider.initialize() then rejects the valid existing table with a dimension-mismatch error. Use the discovered or persisted dimensions for incremental validation rather than treating an unknown model's guessed size as authoritative.

Useful? React with 👍 / 👎.

greptile-apps · 2026-10-30T22:30:42Z

Confidence Score: 3/5

[Medium impact] The PR changes the embedding route used by indexing and semantic search.

Fix custom-model refresh and provider-specific endpoint defaults before merging.

Findings

Custom models cannot refresh ▶
Search calls the wrong server ▶
Stuck requests stall indexing ▶

Summary

Adds native Ollama embeddings through /api/embed, with provider-specific defaults, model-size hints, HTTP mocks, and setup documentation.

Ollama now creates embeddings for indexing and semantic search.
Embedding settings now choose provider-specific models and API hosts.

Diagram

sequenceDiagram
  participant Caller as Indexer or search
  participant Factory as Embedding factory
  participant Ollama as Ollama server
  participant Store as LanceDB
  Caller->>Factory: Resolve provider and model
  Factory->>Ollama: POST /api/show
  Ollama-->>Factory: Model details
  Factory-->>Caller: Ollama provider
  Caller->>Ollama: POST /api/embed
  Ollama-->>Caller: Vectors
  Caller->>Store: Store or search vectors

_{Reviews (1) · Last reviewed commit: "feat(embeddings): add native Ollama embe..." · Reviewed by Greptile}

greptile-apps · 2026-10-30T22:30:45Z

    config.provider ?? parseEmbeddingProviderName(process.env.EMBEDDING_PROVIDER) ?? 'transformers';
  const model = config.model ?? process.env.EMBEDDING_MODEL ?? DEFAULT_MODEL;
  if (provider === 'openai') return model.includes('large') ? 3072 : 1536; // text-embedding-3-large: 3072, all others: 1536
+  if (provider === 'ollama') return getOllamaModelDimensions(model);


Custom models cannot refresh

getConfiguredDimensions keeps using a fixed guess after Ollama returns the actual vector size. If a custom model returns 1024 values and OLLAMA_EMBEDDING_DIMENSIONS is unset, a full build succeeds. A later incremental refresh compares those stored vectors against 768 and throws IndexCorruptedError.

Use the resolved vector size for the incremental check, or require an explicit size before building an unknown model.

Knowledge Base Used: Index storage and integrity

greptile-apps · 2026-10-30T22:30:45Z

  maxRetries: 3,
-  apiKey: process.env.OPENAI_API_KEY
+  apiKey: process.env.OPENAI_API_KEY,
+  apiEndpoint: resolveDefaultApiEndpoint(defaultProvider)


Search calls the wrong server

Provider overrides inherit the endpoint chosen for the environment's provider. If an Ollama index is opened while EMBEDDING_PROVIDER=openai, search supplies the stored Ollama provider and model but retains OpenAI's default endpoint. The new Ollama branch then sends /api/show to OpenAI, so semantic search cannot start.

Resolve the default endpoint from the final provider after merging the caller's settings. Apply the same ordering to the model default.

greptile-apps · 2026-10-30T22:30:45Z

+      const response = await fetch(`${this.host}/api/embed`, {
+        method: 'POST',
+        headers: { 'Content-Type': 'application/json' },
+        body: JSON.stringify({
+          model: this.modelName,
+          input: texts.length === 1 ? texts[0] : texts
+        })
+      });


Stuck requests stall indexing

Neither new HTTP request has an application timeout. If Ollama stops answering while loading or running a model, indexing keeps waiting rather than reporting a useful failure.

Add a configurable deadline to both /api/show and /api/embed, with enough time for a cold model load. This is a non-blocking improvement.

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

chatgpt-codex-connector Bot reviewed Oct 30, 2026

View reviewed changes

greptile-apps Bot reviewed Oct 30, 2026

View reviewed changes

This was referenced Oct 30, 2026

fix: issue triage quick wins (chunk cap, map, path search) #154

Open

Integration preview: 2.3.0 relaunch (do not merge) #158

Draft

Navigation Menu

Repository navigation

feat(embeddings): native Ollama embedding provider - #146

feat(embeddings): native Ollama embedding provider#146
PatrickSys wants to merge 1 commit into
masterfrom
cursor/native-ollama-provider-cdd7

PatrickSys commented Oct 30, 2026

Uh oh!

chatgpt-codex-connector Bot commented Oct 30, 2026 •

edited

Loading

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

chatgpt-codex-connector Bot Oct 30, 2026

Uh oh!

chatgpt-codex-connector Bot Oct 30, 2026

Uh oh!

greptile-apps Bot commented Oct 30, 2026 •

edited

Loading

Uh oh!

greptile-apps Bot Oct 30, 2026

Uh oh!

greptile-apps Bot Oct 30, 2026

Uh oh!

greptile-apps Bot Oct 30, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

PatrickSys commented Oct 30, 2026

Uh oh!

chatgpt-codex-connector Bot commented Oct 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codex Review Summary

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Oct 30, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot Oct 30, 2026

Choose a reason for hiding this comment

Uh oh!

greptile-apps Bot commented Oct 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Confidence Score: 3/5

Findings

Summary

Diagram

Uh oh!

greptile-apps Bot Oct 30, 2026

Choose a reason for hiding this comment

Uh oh!

greptile-apps Bot Oct 30, 2026

Choose a reason for hiding this comment

Uh oh!

greptile-apps Bot Oct 30, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

chatgpt-codex-connector Bot commented Oct 30, 2026 •

edited

Loading

greptile-apps Bot commented Oct 30, 2026 •

edited

Loading