Pull Request Explorer

Summary

Add daily token-usage mode to activity heatmap

Users can now toggle between message counts and token usage in the activity heatmap, giving them clearer insight into daily token consumption.

Health Assessment

• • The PR was reviewed and merged within 2.6 hours with no rework, indicating a smooth, low-friction process.

Summary

refactor: migrate modals to @lobehub/ui/base-ui

Refactors modal components to use the new base-ui imperative modal API, improving consistency and reducing duplication across the application.

Health Assessment

• • The PR was reviewed and merged within 1.5 hours, indicating a smooth process. Despite a large diff (~5,600 lines added/deleted) and 20 files changed, the single review comment and quick merge suggest minimal friction and low risk.

Summary

Add locale keys for limited offer top-up

Adds new locale keys to support limited‑time offer display for free users on the top‑up page, enhancing user experience.

Health Assessment

• • PR merged within minutes with only two small locale files added and no rework required.

Summary

Add skill priority and server runtime for agent builder

Enhances agent builder by prioritizing its tools over LobeHub skills and adding a server runtime to support background execution, improving reliability and configuration accuracy.

Health Assessment

• • Fast cycle time and minimal review comments indicate smooth integration and low risk of regressions.

Summary

Add test for ollamacloud context window error classification

Adds a regression test ensuring ollamacloud context window overflow is correctly classified as ExceededContextWindow, preventing future regressions.

Health Assessment

• • Rapid merge with minimal review, indicating straightforward regression test addition.

Summary

Add Codex & Claude Code guides; merge image/video

Documentation updates provide new bilingual guides for Codex and Claude Code agents and unify image and video generation instructions, improving user onboarding and reducing support queries.

Health Assessment

• • Rapid merge with minimal review indicates straightforward docs update; no code changes reduce risk.

Summary

Fix FK violations as ConversationParentMissing

Eliminates database error noise when users delete topics during agent operations, improving reliability and user experience.

Health Assessment

• • Fast turnaround with a single commit and no rework, indicating a straightforward bug fix with minimal impact on the codebase.

Summary

chore: tighten skill descriptions for triggering

Standardizes skill descriptions to improve trigger accuracy and developer readability.

Health Assessment

• • Fast review and merge with minimal changes; purely documentation update.

Summary

Resolve CLI tools from shell PATH

Ensures the desktop application can locate CLI tools by reading the user’s shell PATH when Electron’s PATH fails, improving tool detection on macOS and Linux.

Health Assessment

• • The PR was reviewed immediately and required only one additional commit, indicating a smooth review process and minimal friction. The change is small to medium in scope and does not introduce significant risk.

Summary

Add execution-device switcher for all agents

Provides users with a device selector for all agent types, enhancing flexibility and adding a desktop app download link.

Health Assessment

• • Merged within 24 hours after a single review, indicating a straightforward change with minimal friction.

Summary

Refactor modal to base-ui imperative API

Switches the custom model modal to an imperative API, simplifying state management and improving consistency across the UI.

Health Assessment

• • The PR was reviewed and merged within 30 minutes, indicating a straightforward change with minimal friction.

Summary

Remove internal Linear issue markers from code

Clean up internal issue references to keep open‑source code clean and avoid exposing internal tracking IDs.

Health Assessment

• • Quick, single‑commit cleanup with no functional changes; minimal review friction and negligible risk to production.

Summary

Refactor modals to imperative API

Simplifies modal invocation by removing state plumbing, improving maintainability and developer experience without affecting user-facing behavior.

Health Assessment

• • Fast cycle time (0.6h) and zero review rounds indicate a straightforward refactor with minimal friction; the large diff size reflects extensive internal restructuring but no breaking changes.

Summary

Restore editor focus after file picker closes

Ensures users can continue typing after uploading files, improving chat usability.

Health Assessment

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

Summary

Clarify local command session handling

Improves agent's ability to accurately interpret local command execution, reducing misinterpretation of command results and handling remote device errors more gracefully.

Health Assessment

• • Quick review and merge with minimal changes suggest low complexity and risk.

Summary

Bump CLI dependency to 0.0.24

Updates the CLI tool version and regenerates the man page, ensuring users see the correct version string.

Health Assessment

• • Fast merge with minimal changes and no review friction.

Summary

Fix skill dropdown close before navigation

Ensures skill settings navigation closes the dropdown, preventing UI glitches and improving user experience.

Health Assessment

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

Summary

Remove markdown syntax from fallback titles

Ensures user-facing titles are clean by stripping markdown before display, improving readability.

Health Assessment

• • Quick fix with no rework, fast review and merge.

Summary

Style sidebar: group spacer with recents and agents

Improves sidebar layout by grouping the spacer with recents and agents, preserving order and independent behavior for a cleaner UI.

Health Assessment

• • The PR was reviewed and merged within 11.5 hours with no rework, indicating a smooth, low-risk style change.

Summary

Refactor topic queries to drop legacy session compatibility

Simplifies the topic data model by removing legacy session-to-agent mapping, improves performance, and updates assistant ranking to be agent-native while preserving runtime backfills for legacy data.

Health Assessment

• • Rapid iteration with AI assistance; quick review and merge within 9 hours; high line changes but minimal review friction.

Summary

Skip user count on API key checks

Improves performance by skipping user count on API key checks.

Health Assessment

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

Summary

Add restore subscription copy

Adds default and zh-CN subscription copy for cloud restore-subscription flow.

Health Assessment

• • Quick review and merge with minimal changes.

Summary

Fix home starter loading stability

Improves reliability of home page starter model loading, reducing errors and enhancing user experience.

Health Assessment

• • Rapid review and merge within an hour, minimal rework, indicates straightforward bug fix.

Summary

Add group-by-status mode to topic sidebar

Enables users to organize topics by status, improving visibility of high-priority items and ensuring critical topics surface on the first page.

Health Assessment

• • Fast turnaround with AI-assisted review, minimal rework, and quick merge within 9 hours.

Summary

Add configurable model routing and starters

Enables customizable model routing and starters for improved user experience.

Health Assessment

• • The PR was merged quickly, but the large scope and lack of human review may indicate potential risks.

Summary

Health Assessment

Summary

Pin Vitest 3.2.4 for deterministic installs

Ensures consistent test environment by locking Vitest and related packages to a specific version, preventing registry drift.

Health Assessment

• • Quick, single-commit PR with minimal changes, merged within 30 minutes, indicating low complexity and high confidence.

Summary

Fix double-encoded array fields in agent creation

Prevents corrupted data in the database caused by LLMs sending array fields as JSON strings, ensuring reliable agent creation and downstream ETL processes.

Health Assessment

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

Summary

Fix default provider setup in business mode

Ensures the LobeHub provider defaults correctly when business features are enabled, improving reliability of AI provider configuration for users.

Health Assessment

• • The PR was merged within 1.1 hours with minimal review comments and a single rework commit, indicating a smooth, low‑friction process.

Summary

Add MiniMax M3 Anthropic video runtime support

Enables MiniMax M3 to use Anthropic’s video API, expanding video content handling for users and improving compatibility with Anthropic-compatible workflows.

Health Assessment

• • Rapid merge with minimal review, indicating high confidence in the changes and low perceived risk.

Summary

Prevent duplicate streaming from stale reconnects

Fixes a race condition that caused duplicate assistant streams, improving reliability and user experience during agent conversations.

Health Assessment

• • Fast turnaround with AI-assisted review enabled a quick, low‑risk fix that addressed a subtle concurrency bug.

Summary

Fix CLI auto-register device on login

Aligns CLI login with desktop by registering the device immediately, ensuring a device record exists on the server and improving user onboarding.

Health Assessment

• • The PR was reviewed and merged within 30 minutes, indicating a straightforward, low‑risk change with minimal friction and no new tests required.

Summary

Update test fixtures for HeterogeneousPersistenceHandler

Align test fixtures with real adapter output to reflect reality and avoid misleading future reviewers.

Health Assessment

• • This PR only updates test fixtures and does not change production code, reducing the risk of introducing new bugs.

Summary

Maintain denormalized topic usage/cost rollup

Adds live aggregation of topic usage and cost metrics from messages, improving analytics and billing accuracy.

Health Assessment

• • The PR was reviewed and merged within 1.6 hours with minimal comments, indicating a smooth process and strong test coverage.

Summary

Register self-iteration builtin tool package

Adds a declarative tool surface for dormant self‑iteration agents, enabling future tool integration without changing existing runtime logic.

Health Assessment

• • Fast 1.6‑hour cycle with no review rounds; large code addition but no production path touched, indicating low risk and high confidence in stability.

Summary

Fix desktop renderer crash by moving visual-ref helpers

This PR resolves a renderer crash on desktop startup by relocating runtime helper functions from a type‑only package to a runtime package, ensuring proper module resolution and preventing syntax errors.

Health Assessment

• • Fast turnaround with minimal changes; quick review and merge indicate low complexity and high confidence.

Summary

Preserve content/state across gateway tool calls

Fixes remote-device tool calls to preserve structured state and avoid leaking success flag into prompt, improving reliability of remote tool execution.

Health Assessment

• • The PR fixed a critical bug in remote tool execution, required significant refactoring across multiple modules, and took nearly 10 days to merge, indicating a complex change with moderate review friction.

Summary

Fix agent builder ID sync to chatStore

Ensures the Agent Builder uses the correct agent context in LLM prompts, improving accuracy and user experience.

Health Assessment

• • Quick, single-commit fix with no blockers; fast review and merge cycle.

Summary

Restore 3 mode‑specific self‑iteration agent slugs

Separates self‑iteration logic into distinct mode‑specific agents, improving routing and future extensibility without affecting current functionality.

Health Assessment

• • Fast review with a single commit, moderate scope refactor that does not alter existing behavior.

Summary

Add May 26 weekly changelog

Adds the latest weekly changelog entry for May 26, keeping documentation up to date for users and developers.

Health Assessment

• • The PR was authored by an AI tool, reviewed immediately, and merged after a single review comment, indicating a smooth process.

Summary

Add PushChannel, receipt cron, and pushToken API

Introduces mobile push notification infrastructure, allowing users to register devices, receive alerts, and manage tokens, enhancing engagement and real‑time communication.

Health Assessment

• • PR merged in under 6 days with a single commit and no major review iterations, indicating a smooth integration.

Summary

Refactor agent runtime to persist nested usage/performance

Aligns metadata persistence across server paths, enabling consistent analytics and future backfill without schema changes.

Health Assessment

• • Merged within 1.5 hours with no rework; minimal code changes and low complexity.

Summary

Add intern-s2-preview support and thinking_mode

Adds support for the intern-s2-preview model and introduces a thinking_mode flag for intern-s series models, improving model selection and stream handling. This enhances flexibility for users and streamlines model configuration.

Health Assessment

• • The PR required multiple commits after an initial review, indicating some rework, but the review was immediate and the overall scope was moderate.

Summary

Add CLI trigger command for agent signals

Enables local testing of agent signal events via CLI, reducing wait times for scheduled tasks and improving developer workflow.

Health Assessment

• • Fast cycle time and minimal review iterations indicate an efficient process, though the large code addition suggests moderate risk for integration testing.

Summary

Add file and image attachments to tasks

Enables users to attach files and images to task interfaces, enhancing collaboration and providing multimodal context for agents.

Health Assessment

• • Large feature with extensive code changes and multiple AI-assisted commits; quick initial review but long cycle time suggests extended testing or integration work was required.

Summary

Fix empty LLM completions retry logic

Ensures that empty LLM responses trigger retries instead of silent completion, improving reliability and user experience.

Health Assessment

• • Fast turnaround with minimal rework; automated review helped catch issues early.

Summary

Fix hardcoded session context values

Replaces static session context placeholders with dynamic template variables, ensuring accurate user, date, and sandbox information is displayed and credentials are correctly injected.

Health Assessment

• • Fast 1.7‑hour cycle and zero‑hour review time indicate minimal friction; the small scope and single commit suggest a straightforward bug fix with low risk.

Summary

Add recent directory management with drag-to-reorder

Allows users to add, reorder, and persist recent directories in device settings, streamlining workflow and eliminating the need for a separate save action.

Health Assessment

• • Fast review and merge with minimal changes indicates a straightforward enhancement.

Summary

Add step-3.7-flash support

Adds support for the new step-3.7-flash model, enabling users to select it in the UI and configure reasoning effort.

Health Assessment

• • The PR was reviewed quickly with automated bot comments and merged within 2.5 days, indicating a smooth process and low risk.

Summary

Add device execution for remote CC with context

Enables heterogeneous agents to run on configured devices, ensuring correct working directory, system context, and online guard, improving reliability and user experience.

Health Assessment

• • Rapid review and merge within 16 hours, indicating high confidence and minimal friction. The large scope was handled efficiently, suggesting a well‑defined feature and effective collaboration.