Pull Request Explorer

Summary

Health Assessment

Summary

Health Assessment

Summary

Health Assessment

Summary

Cache macOS MLX payloads to limit parallel builds

This change limits parallel builds on small runners and caches MLX payloads, improving CI efficiency and reducing build times.

Health Assessment

• • Fast review and merge, minimal changes, straightforward CI improvement.

Summary

Add Ollama model selection to ChatGPT Desktop

Adds Ollama model selection to ChatGPT Desktop on macOS, enabling users to continue chats with configured plugins, MCPs, and skills.

Health Assessment

• • Rapid review and merge (under 1 hour) suggest high confidence in the changes, while the large code churn indicates a substantial new feature addition.

Summary

Add Codex compaction endpoint for OpenAI

Adds an OpenAI compatible compaction endpoint

Health Assessment

• • The PR was merged within 20 hours with a single review, indicating a smooth and efficient implementation process.

Summary

Add client tool search support

Enables clients to search for tools in OpenAI responses, improving developer experience and reducing lookup time.

Health Assessment

• • The PR introduced a substantial number of lines across three files, indicating a significant feature addition. The review process took almost a full day, suggesting moderate complexity but no major blockers.

Summary

Enable speculative decoding for structured output

Adds speculative decoding support for structured output, improving throughput by 4‑8% on large models while ensuring grammar compliance.

Health Assessment

• • The PR had a long cycle time of 142 hours and no early review, indicating high complexity and potential integration risk.

Summary

Rebuild MLX macOS test payloads

Ensures MLX unit tests run correctly on macOS by rebuilding stale payloads when dependencies change.

Health Assessment

• • Fast review and merge with minimal rework; low risk and moderate scope.

Summary

Add image and audio support to gemma4

Gemma4 models can now process image and audio inputs, broadening application scenarios. This unlocks richer multimodal interactions without requiring new imports.

Health Assessment

• • The PR introduced extensive new functionality for multimodal input, requiring significant code additions across multiple modules. Despite a moderate review time, the large scope and lack of multiple review rounds suggest a high risk of integration issues.

Summary

Add comprehensive error checking to MLX bindings

Ensures MLX operations fail fast with clear error messages, preventing silent failures and downstream crashes.

Health Assessment

• • The PR took over two weeks to merge with a single review round, indicating a slow review process. The large code changes across many files suggest significant refactoring and potential risk.

Summary

Fix MTP test alignment in mlx

Corrects unit test misalignment, ensuring new helpers are properly wired.

Health Assessment

• • Quick fix with minimal changes, no rework needed.

Summary

Update llama.cpp version to b10760

Bumps llama.cpp to the latest version, ensuring compatibility and bug fixes.

Health Assessment

• • Quick version bump with minimal changes, fast review and merge.

Summary

Add cached prompt token metrics

Enables accurate measurement of cache usage, improving performance insights and cost optimization for users.

Health Assessment

• • The PR required over 10 days to merge with a single review cycle, indicating a slow review process and potential bottlenecks in the workflow.

Summary

Update llama.cpp to version b10729

Bumps llama.cpp to the latest version and regenerates compatibility hooks to support new tensor loading behavior, ensuring seamless integration with Ollama's backend.

Health Assessment

• • PR merged within 24 hours after review with minimal rework, indicating a smooth integration process.

Summary

Update MLX and MLX-C version

Updates MLX and MLX-C to newer versions, removing an obsolete patch. This ensures compatibility with the latest releases and simplifies maintenance.

Health Assessment

• • The PR had a long review cycle and required multiple commits to address comments, indicating significant review friction. The large code changes across backend and devops files suggest a high-risk update.

Summary

Build Go Dependencies

Adds Go dependency management and license generation to the build pipeline, ensuring all Go modules are properly licensed and included in releases.

Health Assessment

• • Fast review and a single rework commit indicate a smooth, low‑risk process.

Summary

Deduplicate dependency files

Remove duplicate license files from runner directories and package them in base packages, simplifying build and reducing redundancy.

Health Assessment

• • Quick review and merge indicates a straightforward refactor with minimal impact.

Summary

Health Assessment

Summary

Add account cloud models to Claude picker

Enables users to see all signed‑in cloud models in the Claude Desktop picker, improving model selection flexibility and ensuring defaults are driven by recommendations.

Health Assessment

• • Quick turnaround with minimal rework; PR merged within 20 minutes of opening.

Summary

Synchronize macOS app handoff

Ensures only one Ollama app instance runs after updates, improving reliability and user experience by preventing stale processes and race conditions.

Health Assessment

• • Fast cycle time and short review window indicate efficient collaboration; multiple commits refined handoff logic, but the large line change set suggests moderate complexity.

Summary

Continue requests across catalog refreshes

This change eliminates intermittent errors when the Claude model catalog changes, ensuring smoother user experience and reducing retry overhead.

Health Assessment

• • Rapid review and merge indicate low complexity and high confidence in the fix.

Summary

Restore system dark mode

Users can now experience consistent light and dark appearance across the desktop UI and Apps Launcher, improving usability and visual consistency.

Health Assessment

• • PR completed quickly with minimal iterations, indicating a straightforward implementation.

Summary

Consume Claude recommendation mappings

Integrates the new Claude recommendation API to provide more accurate model selections, improving user experience and reducing misconfigurations.

Health Assessment

• • The PR was merged within 4.7 hours, indicating efficient review and low friction.
• • Despite a large code change, the short cycle time suggests high confidence and minimal risk.

Summary

Remove text-only Gemma3 model from MLX

Eliminates unsupported Gemma3 text‑only model, ensuring correct engine routing and reducing maintenance overhead.

Health Assessment

• • Quick turnaround with a single review and commit, indicating a straightforward cleanup of unused code.

Summary

Fix Pi host configuration to respect OLLAMA_HOST

Ensures Pi connects to the correct Ollama server when OLLAMA_HOST changes, preventing stale localhost connections and improving reliability for remote deployments.

Health Assessment

• • Quick fix with minimal changes, resolved in a single review cycle, indicating low complexity and high confidence in the patch.

Summary

Prevent model picker scroll jumps

Improves the settings UI by eliminating jarring scroll behavior when selecting Claude models, enhancing user experience and reducing confusion.

Health Assessment

• • Fast turnaround with minimal changes; low complexity and risk.

Summary

Fix llama.cpp Docker build on Linux

Resolved missing cmake utility in Docker build context, ensuring successful image creation for Linux environments.

Health Assessment

• • PR merged immediately with no review, indicating a straightforward and low-risk fix.

Summary

Health Assessment

Summary

Add Claude Desktop Integration Docs

Provides updated documentation for Claude Desktop integration, improving user onboarding.

Health Assessment

• • Rapid merge with minimal changes indicates a straightforward documentation update.

Summary

Add Claude Desktop integration documentation

Provides updated setup instructions and feature overview for Claude Desktop, improving user onboarding.

Health Assessment

• • Quick documentation update with minimal changes and fast review, indicating low complexity and high efficiency.