Pull Request Explorer

Summary

Fix per-agent working directory store handling

Ensures agent working directories are stored in state, enabling reactive updates and consistent UI across components.

Health Assessment

• • Fast turnaround with minimal changes and quick review, indicating low complexity.

Summary

Fix heavy fields in agent runtime stream events

Reduces Redis write size to avoid exceeding limits, preventing runtime crashes and improving system reliability for agent operations.

Health Assessment

• • Fast turnaround with a single commit and no review rework indicates a straightforward, low‑risk bug fix that quickly restored stability to the agent runtime stream handling.

Summary

Fix heteroIngest race by passing assistantMessageId

This change eliminates race conditions that caused agent executions to fail, ensuring conversations complete without errors. It improves reliability and user experience for agents.

Health Assessment

• • Quick resolution with minimal review, indicating a straightforward bug fix.

Summary

Improve Vite route chunk preloading

Enhance performance by prewarming more route chunks during idle time, covering additional desktop secondary routes.

Health Assessment

• • The PR was merged quickly with minimal review comments, indicating a straightforward and well-tested change.

Summary

Add missing i18n keys for follow-up feature

Ensures Settings UI displays localized labels for follow-up suggestions in Chinese and English, improving user experience.

Health Assessment

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

Summary

Fix llm-generation-tracing scenario backfill

Corrects tracing metadata so new rows record proper task brief and judge scenarios, improving data accuracy and reducing unknown entries.

Health Assessment

• • The PR was merged within 36 minutes with no review rounds, indicating a straightforward, low‑risk fix that required minimal code changes.

Summary

Fix Turbopack glob warning in LLM tracing

This change removes a noisy build-time warning that was inflating the build trace. It improves developer experience by keeping build logs clean and reducing build time.

Health Assessment

• • Quick fix with minimal changes, no rework, fast review, low risk.

Summary

Implement Hermes agent chain logic

Adds direct CLI integration for Hermes agents, enabling faster, multi‑turn chat responses and improved platform capability for users.

Health Assessment

• • Fast review cycle (0h) and single round of AI‑powered feedback led to a quick merge with minimal friction.

Summary

Fix image responses to explain text-only content

Improves user experience by providing clear guidance when image generation returns text-only responses, preventing confusion and mislabeling safety refusals.

Health Assessment

• • The PR was merged within 12 hours with minimal review comments, indicating a smooth review process and low complexity.

Summary

Add tracing ID pre-allocation and feedback router

Enables per-call feedback loop for LLM generation, improving observability and user feedback handling.

Health Assessment

• • Fast review and single commit indicate smooth process, but large scope and many files suggest high impact and potential risk.

Summary

Normalize, dedupe, and fix project overview

Streamlines skill documentation by removing duplicate entries and standardizing descriptions, improving maintainability. Adds an audit skill to automate future checks, ensuring consistent quality.

Health Assessment

• • Rapid review and merge with minimal iterations, indicating straightforward documentation updates.

Summary

Add task-handoff tracing and prompt versioning

Adds precise task-handoff tracing and prompt versioning, improving analytics and debugging. Also updates UI preview styling for clearer file previews.

Health Assessment

• • Quick turnaround with minimal rework; AI assistance accelerated development.

Summary

Add execution target switcher in chat composer

Enables users to select where heterogeneous agents run—desktop, cloud sandbox, or remote device—enhancing flexibility and control over agent execution.

Health Assessment

• • Rapid review and single commit cycle, but the PR introduces a large, cross‑cutting feature spanning frontend UI and backend dispatch logic, indicating a high‑impact change with potential integration complexity.

Summary

Add execAgent plumbing for self-iteration migration

Establishes infrastructure for future agent execution enhancements, enabling safer migration without impacting current functionality.

Health Assessment

• • PR merged within 30 minutes with minimal review, indicating low complexity and high confidence in the changes.

Summary

Health Assessment

Summary

Refactor user memories to resolve agent config

This refactor enables user memory services to resolve agent configuration from per‑user ServiceModel settings, improving flexibility and reducing reliance on environment defaults.

Health Assessment

• • The PR had a quick review turnaround and minimal back‑and‑forth, but the large code churn suggests significant refactoring effort.

Summary

Fix image MIME normalization for uploads

Ensures image MIME types match actual bytes, preventing provider validation errors and improving upload reliability.

Health Assessment

• • Fast 1.5‑hour cycle with AI‑driven reviews and no human blockers, quickly resolving MIME mismatch issues.

Summary

Fix agent projectSkills visibility regardless of device

Ensures client-provided project skills are always surfaced to the agent, preventing silent failures when device resolution fails. This improves reliability and user experience.

Health Assessment

• • Quick fix with minimal changes, no rework, fast review, low risk.

Summary

Fix Bedrock structured generation support

Adds support for Bedrock structured generation in the model runtime, fixing production failures.

Health Assessment

• • The PR was merged quickly, but the scope of changes is large, indicating potential risk.

Summary

Fix agent runtime to preserve streamed content

Ensures partial assistant responses are retained when a stream is cancelled mid‑stream, preventing data loss and improving user experience.

Health Assessment

• • Fast cycle time and minimal review comments indicate a smooth, low‑risk bugfix process.

Summary

Clean up LOBE-xx comment references

Automated removal of placeholder comments and added architecture context, improving code clarity.

Health Assessment

• • Quick review and single commit indicate minimal impact and low risk.

Summary

Fix desktop preview for unknown file extensions

Ensures that files such as .cjs, .mjs, and config files are correctly previewed as text instead of being misidentified as binary, improving user experience in the desktop app.

Health Assessment

• • Quick fix with minimal review, AI-assisted code generation, resolved preview issue promptly.

Summary

Refactor agent runtime to use device gateway

This refactor consolidates desktop, web, and bot callers onto a single device-gateway dispatch path, removing legacy desktop shortcuts. It reduces bugs and ensures consistent template variable handling across platforms.

Health Assessment

• • Quick turnaround with a single commit and immediate approval indicates low complexity and minimal rework.

Summary

Add Qwen3.7 Max locale description

Adds locale description for Qwen3.7 Max, enabling accurate model card copy rendering for LobeHub-hosted models.

Health Assessment

• • Quick approval with no rework, minimal changes, low risk.

Summary

Normalize block spacing in assistant workflow

Improves visual consistency of workflow content blocks, reducing UI clutter and enhancing user experience.

Health Assessment

• • Quick style fix with minimal changes, reviewed and merged within an hour, indicating smooth process.

Summary

Consume gateway uiMessages snapshot as SoT

Client now uses server‑pushed UIChatMessage snapshot as source of truth, eliminating refetches and preventing stale data during streaming.

Health Assessment

• • Rapid AI‑assisted development with minimal review friction, but large scope and many commits suggest high complexity and potential risk.

Summary

Hide community publish for heterogeneous agents

Hides the assistant profile 'Publish to Community' action for agents backed by heterogeneous runtimes.

Health Assessment

• • The PR was merged quickly with minimal review time, indicating a straightforward fix.

Summary

Fix agent settings defaults and locale state

Ensures new agents load with correct defaults and UI text is properly localized, preventing user confusion and improving agent configuration reliability.

Health Assessment

• • The PR was merged within an hour of opening, with no rework or additional review rounds, indicating a straightforward, low‑risk fix.

Summary

Fix model name swap for remote hetero agents

Ensures local CLI heterogeneous agents display the actual model ID instead of the provider brand label, improving clarity for developers.

Health Assessment

• • Quick fix with minimal changes, AI-assisted code generation, and immediate approval.

Summary

Replace desktop gateway agent run flow with direct lh hetero exec

This PR fixes the agent run flow on desktop by replacing the previous multi‑step pipeline with a direct hetero exec spawn, ensuring proper communication with the server and eliminating duplicate token errors.

Health Assessment

• • Quick fix with minimal changes, AI‑assisted review, no blockers, merged within a day.

Summary

Add UIChatMessage snapshot at gateway boundaries

Enables clients to use server‑pushed UIChatMessage snapshots instead of refetching from DB, reducing stale data and improving streaming consistency.

Health Assessment

• • Fast review and merge with minimal friction; quick iteration to address stale data issue.

Summary

Fix agent-runtime context engine payload routing

Reduces heavy context engine payload from per-step events, improving state storage efficiency and trace accuracy. This change cleans up the event stream, ensuring only relevant trace data is persisted.

Health Assessment

• • Fast turnaround with minimal rework; AI-assisted code generation streamlined the fix.

Summary

Group review changes by submodule

Enhance the review panel to cluster changes by submodule, improving user experience for those working with submodules.

Health Assessment

• • The PR has a large scope with many files changed, but the review process was relatively smooth and quick.

Summary

Add per-call LLM generation tracing observability

Enables detailed per-call observability for LLM generateObject calls, storing trace data in a database and optional S3, improving monitoring and debugging.

Health Assessment

• • Rapid single-commit merge with zero review delay indicates low friction, but the large code addition (~2.9k lines) and extensive feature scope raise overall risk.

Summary

Clean up LOBE-XXX code annotations

Removes legacy LOBE-XXX markers from comments, improving code readability and preventing test syntax errors.

Health Assessment

• • Quick turnaround with minimal rework; automated reviews flagged issues but were resolved swiftly.

Summary

Add multi-select delete in document tree

Enables users to delete multiple documents at once, improving productivity and reducing errors.

Health Assessment

• • Quick turnaround with minimal review, indicating straightforward feature addition.

Summary

Translate non-English comments to English

Non-English comments in agent skill identifiers were translated to English, improving code readability and maintainability.

Health Assessment

• • Quick turnaround with minimal review, indicating a straightforward change.

Summary

Clean up settings page copy and entries

Improves clarity and usability of the settings page by updating copy, ensuring provider options are always visible, renaming developer mode to advanced tools, and removing duplicate telemetry controls.

Health Assessment

• • Fast cycle time and immediate review indicate an efficient process with minimal rework; the PR was merged within 13.5 hours without additional review rounds.

Summary

Add onboarding analytics event tracking

Enables reliable onboarding funnel tracking, improving user acquisition insights and reducing reliance on SPA pageviews.

Health Assessment

• • Fast cycle time and immediate review indicate high confidence in the changes and minimal risk to the codebase.

Summary

Remove unused PostHog component

Eliminates dead code, reducing bundle size and maintenance overhead.

Health Assessment

• • Quick cleanup with minimal review, no tests required.

Summary

Pin baseline-browser-mapping to 2.10.31

Ensures consistent dependency resolution to avoid Next.js build failures, improving build stability.

Health Assessment

• • Quick fix with minimal changes, merged immediately after review, indicating low complexity and high confidence.

Summary

Improve route chunk preloading performance

Enhances application startup speed by preloading route chunks during idle periods, reducing load times for users.

Health Assessment

• • The PR underwent rapid iteration with multiple commits following an initial review, indicating active refinement. The large code churn and many files suggest a significant performance enhancement effort.

Summary

Fix Docker prepare script for missing git

Ensures Docker builds succeed by guarding against missing git during package installation, improving CI reliability.

Health Assessment

• • Quick fix with minimal changes, AI review provided, merged within 20 minutes.

Summary

Extend follow-up chip suggestions to general chat

Adds cross‑surface follow‑up suggestions, improving user engagement and reducing friction across inbox, 1:1, popup, and portal threads.

Health Assessment

• • The PR was reviewed immediately and merged within 11 hours, indicating a smooth review process, but the large number of commits and extensive code changes suggest a high‑risk, high‑impact feature implementation.

Summary

Add LLM Generation Tracing and Agent Eval Tables

Adds database tables for tracking LLM generation and agent evaluation experiments, enabling richer analytics and debugging for AI workflows.

Health Assessment

• • Merged quickly with minimal review, indicating a straightforward schema merge.

Summary

Refactor: Add Shared Guard Helpers

Centralizes guard utilities across the codebase, reducing duplication and improving maintainability.

Health Assessment

• • Rapid review and merge within 1.1 hours, minimal back‑and‑forth, but large code churn indicates significant refactoring effort.

Summary

Add support for Jina CN domains

Enables users in mainland China to access Jina-powered search by switching to CN domains, improving reliability and user experience.

Health Assessment

• • The PR was merged after a single commit with no review, showing minimal friction, but the 4.5‑day cycle time suggests a delay before the merge was approved.

Summary

Fix cli websocket agent run crash handling

Catches promise error to avoid agent run crash in WS mode, improving overall system reliability.

Health Assessment

• • The PR was reviewed and merged within a reasonable timeframe, indicating a smooth development process.

Summary

Fix Gemini image diagnostics preservation

Ensures Gemini image responses are retained for error handling, improving diagnostics and downstream processing.

Health Assessment

• • The PR was merged within 0.7 hours with minimal review, indicating a straightforward bug fix that was quickly validated by automated AI review tools.

Summary

Show check with warning badge for partial-success runs

Adds a visual cue to indicate partial failures in workflow turns while keeping the overall status as successful, improving user clarity.

Health Assessment

• • Merged within 30 minutes with no review comments, indicating a straightforward style update with minimal risk.