Pull Request Explorer

Summary

Fix laguna patch build breakage

Resolves build failures for the Laguna patch, ensuring stable library exports.

Health Assessment

• • Quick fix with minimal changes, merged within minutes, no review required.

Summary

Ignore SSE ping comments in llama-server

Fixes streaming behavior by skipping non‑data SSE comment frames, ensuring correct JSON parsing for completion and chat streams.

Health Assessment

• • Merged within minutes with a single commit, indicating a straightforward, low‑risk fix.

Summary

docs: configure hermes desktop app

Provides manual steps to set up the Hermes desktop app for use with Ollama, improving user onboarding.

Health Assessment

• • Rapid documentation update with minimal code changes; low risk and quick review.

Summary

Add Laguna architecture support via llama.cpp patch

Adds support for the Laguna architecture to Ollama's llama.cpp integration, enabling new model variants and improving inference performance for certain workloads.

Health Assessment

• • PR was reviewed and merged within a day, indicating smooth integration and minimal friction.

Summary

Isolate Codex launch configuration

This PR creates a dedicated Codex CLI profile, isolates launch configuration, and adds cleanup for CLI‑owned files, improving maintainability and reducing configuration conflicts.

Health Assessment

• • Merged within 18 minutes with a single review, indicating high confidence and low review friction despite the large code changes.

Summary

Health Assessment

Summary

Health Assessment

Summary

Detect llama-server load stalls

Adds logic to monitor llama-server output to prevent timeouts during model loading, improving reliability.

Health Assessment

• • Fast turnaround with minimal rework; single commit resolved the issue.

Summary

Enable Radeon 8060S iGPU by default

Adds ROCm gfx1151 to allowlist, ensuring the Radeon 8060S integrated GPU is enabled without requiring OLLAMA_IGPU_ENABLE, improving GPU compatibility for users.

Health Assessment

• • Quick fix with minimal changes, fast review and merge.

Summary

Health Assessment

Summary

Add cached prompt token counts to llama-server

Preserves legacy prompt evaluation metrics by combining cached and new prompt tokens, ensuring backward compatibility for Ollama users. This change maintains accurate token accounting for existing workflows.

Health Assessment

• • Quick review and merge indicate a straightforward bug fix with minimal impact.

Summary

Fix OpenCode local model limits

Adjust OpenCode behavior to omit limit stanzas for local models lacking output limits, restoring pre‑0.30 functionality.

Health Assessment

• • Quick fix with minimal changes and fast review, indicating low complexity and low risk.

Summary

Add detailed template selection logging

Enables better troubleshooting by logging template selection details, improving debugging and reliability.

Health Assessment

• • Fast review and merge with minimal changes, indicating low complexity and risk.

Summary

Show and auto-install Cline CLI

Adds Cline CLI visibility in the launcher and auto‑install support via npm, with comprehensive tests.

Health Assessment

• • Fast cycle time and minimal rework indicate a straightforward feature addition.

Summary

Migrate Codex configuration for launch

Updates Codex configuration to improve compatibility and support local model metadata and prompt arguments.

Health Assessment

• • Quick turnaround with minimal changes indicates a straightforward bug fix with low risk.

Summary

Fix Cline Ollama provider configuration

Ensures Cline CLI correctly configures the Ollama provider, enabling seamless local and cloud model usage while preserving legacy compatibility.

Health Assessment

• • Long review cycle and large code changes suggest significant rework; no AI involvement; risk high due to potential impact on provider configuration.

Summary

Avoid legacy Codex App profiles

Prevents launch failures by removing legacy profile settings, ensuring compatibility with newer Codex versions and maintaining user-selected provider configuration.

Health Assessment

• • The PR required a single review after a long review cycle, indicating moderate review friction but overall straightforward changes.

Summary

Update llama server and CI for llama.cpp

Adds multi‑GPU support, fixes embeddings API, updates CI for llama.cpp, and resolves imagegen loading bug, improving stability and build reliability.

Health Assessment

• • PR involved extensive code changes across backend and CI, with a slow initial review and multiple commits, indicating significant rework and potential risk.

Summary

Fix Qwen3Next model conversion issues

This PR resolves conversion bugs for Qwen3Next models, ensuring correct tensor naming and quantization, improving model compatibility and performance.

Health Assessment

• • The PR fixed conversion bugs quickly with minimal rework, indicating efficient code review and clear issue resolution.

Summary

Health Assessment

Summary

Health Assessment

Summary

Fix MLX dev mode library search path

Corrects library resolution paths after superbuild changes, ensuring MLX runs correctly in development mode.

Health Assessment

• • Quick fix with minimal changes, fast review and merge.

Summary

Fix MLX show API information

Corrects model metadata display, improving accuracy for users.

Health Assessment

• • Quick fix with minimal changes, fast review and merge.

Summary

Remove duplicate template parsing

Eliminates redundant template parsing in server code, preserving behavior and adding focused unit tests.

Health Assessment

• • Quick turnaround with minimal changes and a single review round indicates a straightforward refactor with low risk.

Summary

Revert DFlash speculative decoding integration

Reverts an invasive DFlash integration to keep the codebase clean and reduce complexity.

Health Assessment

• • The PR was merged quickly with minimal review, indicating low friction, but the large code changes and removal of a complex feature suggest moderate risk.

Summary

Add enriched model inventory to launch

Enables richer model metadata for launch, reducing API calls and improving integration efficiency.

Health Assessment

• • PR required a second commit after review, indicating some rework, but was merged within two days, showing efficient collaboration.

Summary

Health Assessment

Summary

Omit patch tool type from codex

Removing the patch tool type prevents schema compatibility issues, ensuring stable deployments.

Health Assessment

• • Quick fix with minimal changes, resolved in under two hours with no review back‑and‑forth.

Summary

Improve release build speed and efficiency

Optimizes CI release pipeline, reduces build times, improves caching, and lowers resource usage, boosting release velocity and developer experience.

Health Assessment

Summary

Health Assessment

Summary

Health Assessment

Summary

Add DFlash speculative decoding to MLX runner

Enables efficient speculative decoding for DFlash blocks, improving inference speed for Qwen3.6 and related models.

Health Assessment

• • Merged within 19 hours with minimal rework, indicating a streamlined review process.

Summary

Improve codex app restart reliability

Enhances stability of codex app restarts, reducing brittleness and improving user experience.

Health Assessment

• • Quick turnaround with minimal changes; PR merged within 30 minutes, indicating low complexity and high confidence in the fix.

Summary

Health Assessment

Summary

Health Assessment

Summary

Rework MLX sampler distribution pipeline

Improves sampling efficiency and reproducibility by restructuring the MLX sampler pipeline, enabling sparse token handling and fixing seed behavior.

Health Assessment

• • The PR was merged quickly with minimal review, indicating a straightforward bugfix with low complexity.

Summary

Add Codex App Launch Integration

Enables users to launch and manage the Codex App directly from the platform, streamlining workflow and expanding integration options.

Health Assessment

• • The PR required multiple commits and a review after 24 hours, indicating moderate review friction. The large code additions and numerous test files suggest a substantial feature implementation with potential impact on stability.

Summary

Add memory trace logging

Adds memory trace logging to help diagnose issue #16030.

Health Assessment

• • Quick, single-file change with minimal impact; fast review and merge cycle indicates low complexity and risk.

Summary

Add image modalities for vision models

Adds image input capability metadata to OpenCode config for vision‑capable Ollama models, enabling better integration and user experience.

Health Assessment

• • Long review and cycle times suggest process bottlenecks, but the change is small and low risk.

Summary

Preserve Claude local image-path tool results

Ensures images referenced by local paths are correctly handled and displayed in locally served models, improving user experience for image-based prompts.

Health Assessment

• • The PR required extensive rework across multiple renderers and prompt logic, leading to a large code change set and a slow cycle time, but the review process was relatively straightforward with a single round of comments.

Summary

Fix macOS 26 target leakage in metallib

Corrects deployment target leakage in Metal v3 library, ensuring proper macOS compatibility for MLX builds.

Health Assessment

• • Long review time indicates a potential blocker or low priority; the change is small and confined to a single build script, implying low risk to the overall system.

Summary

Avoid status timeout during inference

Improves stability of the /v1/status endpoint by caching VRAM usage and preventing long-running inference from blocking health checks, ensuring reliable server health reporting.

Health Assessment

• • The PR took a full day to review, indicating potential complexity or low priority. The change is a bugfix with moderate scope, but the review process was slow, suggesting limited urgency or resource constraints.

Summary

Update imagegen runner for mlx thread affinity

Enhances image generation by applying mlx thread affinity, improving performance and reliability for users.

Health Assessment

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

Summary

Harden update flows

Adds opt‑in and unit tests to secure Windows and macOS update processes.

Health Assessment

• • Fast review and merge with minimal rework indicates low complexity and high confidence.

Summary

Improve integration test reliability

Enhances test stability on slower systems, reducing flaky failures.

Health Assessment

• • PR focuses on test stability with minimal code changes, likely low risk.

Summary

Disable Claude Desktop launch

Removes integration with Claude Desktop due to blocked third-party access for non-anthropic models.

Health Assessment

• • Fast review time and approval suggest a well-structured and low-risk PR.

Summary

Add plan-aware model gating to launch

Adds plan-aware gating to ollama launch, providing sign‑in or upgrade hints for cloud models, improving user guidance.

Health Assessment

• • Rapid approval with minimal changes indicates low complexity and high confidence.

Summary

Improve integration backup UX

Enhances backup handling for integration configs, keeping only recent backups and removing unnecessary prompts, which streamlines user experience and reduces disk clutter.

Health Assessment

• • The PR had a long cycle time and delayed first review, but limited back‑and‑forth discussion, indicating a slow but straightforward review process.

Summary

Add server-side cache for /api/show responses

Improves model capability lookup speed by caching full responses, reducing latency for launch flows.

Health Assessment

• • Fast cycle time and minimal review comments indicate smooth integration.

Summary

Add Gemma4 MTP speculative decoding support

Enables faster inference by allowing multi-token speculative decoding for Gemma4 models, improving performance and reducing latency for end users.

Health Assessment

• • PR merged quickly with minimal review, indicating a straightforward implementation with low risk.