v2.1.0-ce

✨ New Features

• AI Gateway & OpenAI-Compatible APIs

  • Audio Transcription: Added OpenAI-compatible /v1/audio/transcriptions support with multipart request rewriting and audio token usage counting.
  • Text/Image-to-Video: Added /v1/videos, /v1/videos/{id}, and /v1/videos/{id}/content APIs with provider adapters for OpenAI-compatible endpoints, LightX2V, MiniMax, and Seedance.
  • Model Routing: Added provider-aware model IDs, composite model ID parsing, upstream catalog support, session routing, fallback retry for chat completions, and per-upstream availability reporting.
  • Usage Limits: Added Redis-backed per-window usage limit checks for configured upstream policies.
  • API Key Auth: AI Gateway inference endpoints now require user/org API keys instead of normal login sessions.

• API Key Management

  • Added namespace-scoped API key management for users and organizations, including create, list, update, delete, built-in key retrieval, and built-in key refresh APIs.
  • Added user/org API key authentication context propagation for downstream services.

• Inference & Evaluation

  • Added configurable model architecture checks for inference, including admin APIs to view and update inference architecture rules.
  • Added SGLang-based Qwen3-Guard stream inference configuration and Docker assets.
  • Added AMD EvalScope evaluation configuration and Docker image support.
  • Updated vLLM and AMD vLLM inference images/configuration to v0.19.0.

• Repository, Tags & Skills

  • Added automatic industry tag scanning for model and dataset repositories using configured LLM prompts.
  • Added source tracking for repository tags and safer tag replacement/removal behavior.
  • Added skill mirror_from_saas routes and skill clone URL fields.
  • Added a dedicated skill tag category seed.
  • Improved SKILL.md validation and added broader validator tests.

🚀 Enhancements & Bug Fixes

• AI Gateway Reliability

  • Fixed async model cache writes mutating live model lists.
  • Fixed nil-user panic risk when listing CSGHub models.
  • Improved sensitive-check whitelist lookup behavior.
  • Made sensitive-check behavior configurable per LLM config where available.
  • Improved SGLang Guard stream trace/session header handling.

• Resource Scheduling

  • Added unavailable reasons to resource list responses.
  • Added cluster offline/unavailable status handling.
  • Prevented CPU-only workloads from being scheduled onto XPU nodes.
  • Added replica-aware resource checks for Spaces.

• Repository & LFS

  • Added repository size calculation trigger command.
  • Added LFS pointer download nil-URL protection.
  • Added LFS size checks before syncing files.

• Data Viewer

  • Added file-size checks and optimizations before converting preview files.

• Finetune & Runner

  • Fixed finetune jobs missing model and dataset revision data.
  • Fixed potential runner panic paths in service/workflow handling.

• Proxy & Networking

  • Set proxied Host headers without port where required.
  • Sanitized logged authorization headers in internal proxy logs.

🛠 Maintenance

• Upgraded vulnerable dependencies reported by Dependabot.
• Improved accounting metering retry limit configurability.
• Added and refreshed unit tests across AI Gateway, API keys, resource checks, tags, skills, LFS, and database stores.
• Improved CI/test stability and separated CI build cache behavior.

Full Changelog: v2.0.0-ce...v2.1.0-ce

v2.0.0-ce

✨ New Features

• Repository & Git Operations
  • Branch Management: Added APIs for creating and deleting branches.
  • Resumable LFS: Enabled file-level resumable uploads for large files and implemented MinIO LFS cleanup on model deletion.
  • Performance: Optimized repository query speeds and pre-upload logic for large file counts.
• Scheduling & Resource Management
  • Memory Unit Support: Enhanced resource checks to support "Mi" and "M" units for memory configuration.
  • Knative Integration: Updated service namespaces to knative-serving and improved pod name RFC1123 compliance.
• AI Gateway & LLM Enhancements
  • Multimodal Support: Added Text-to-Image capabilities and improved image generation proxying.
  • Engine Upgrades: Upgraded SGLang to support Qwen 2.5, GLM 4.5, and Step-3; added data-parallel-size config for vLLM.

🚀 Enhancements & Bug Fixes

• Security & Authentication
  • Implemented a JWT session-building middleware.
  • Enabled API Key authentication for Runner APIs.
  • Restricted GET /v1/models to allow unauthenticated access.
• Skills
  • Added comprehensive APIs for Agent skills, enhance tags support
  • Corrected skill tag scope bugs and validated SKILL.md file integrity.
• Observability & Reliability
  • Cluster Metrics: Added cluster heartbeat metrics and automated alerting rules.
  • Enhanced Logging: Integrated Loki for better finetuning logs and improved SMTP attachment handling using in-memory bytes.
  • Stability: Fixed nil deployer panics in MCP proxies and improved wakeup timeout handling for remote clusters.
• Bug Fixes
  • Resolved token usage recording issues for interrupted streaming chat requests.
  • Fixed status mismatch bugs where deleted Knative services still appeared as "Running."
  • Fixed repository total size calculation bugs and search prioritization for organization namespaces.

🛠 Maintenance

• Upgraded Docker Golang version to 1.26.0.
• Refactored metering consumer logic into a unified structure.
• Added extensive unit testing for accounting workflows.

New Contributors

• @sfeng1996 made their first contribution in #878

Full Changelog: v1.17.0-ce...v2.0.0-ce

v1.17.0-ce

v1.17.0-ce

✨ New Features

• Advanced Scheduling with Volcano
  • Integrated Volcano Scheduler to enhance high-performance workload management.
  • Added support for Volcano vGPU resource definitions and MIG (Multi-Instance GPU) resources with full utility and test coverage.
• Infrastructure & Compute Visibility
  • Introduced Public Cluster Info for a dedicated computation power page.
  • Added Cluster Resource Health Check to monitor deployment readiness.
  • Enhanced cluster management APIs to support portal integration and real-time node/queue data collection.
• Deployment & Runtime Enhancements
  • Added Pending Status for deployments to provide better feedback during resource allocation.
  • Introduced RuntimeFrameworkID and DriverVersion fields to deployment and space displays for better environment tracking.
  • Supported updating resource and cluster IDs for existing inference and finetune tasks.
• Agent & Skills Hub
  • Introduced Skills Support with multi-sync capabilities and a "User Likes" API.
  • Added Agent Access Middleware that enables token usage tracking.
• Security & Authentication
  • Added Basic Auth support to the Authenticator.
  • Introduced a configurable bypass for sensitive content detection in the AI Gateway.
  • Enabled access token support for non-MCP spaces via reverse proxy.

🚀 Enhancements & Bug Fixes

• Storage & Git Performance
  • Optimized LFS Sync pointer retrieval and fixed pre-upload timeout issues for repositories with 300k+ files.
  • Fixed MinIO multipart upload errors and addressed metadata EOF issues via storage gateway pre-signed URLs.
  • Added ScanFileNumLimit configuration to prevent overhead during file scanning.
• Reporting & Data Export
  • Added Time Range Queries and CSV Export functionality for system reports.
  • Updated organizations table with a UUID column and implemented conflict checks.
• System Stability
  • Added configurable timeouts for Temporal GetSystemInfo calls.
  • Added a Scaffold Command to streamline code generation for developers.
• Bug Fixes
  • Fixed a data duplication bug in the Models API.
  • Resolved AMD GPU inference issues within Kubernetes (K8s) environments.
  • Fixed xnet migration status filters for models and datasets index APIs.
  • Addressed various error-handling bugs, including non-existent service deletion and MinIO upload failures.

New Contributors

• @sheperdh
• @Boulim
• @awanawana
• @hwzhuhao

Full Changelog: v1.16.1-ce...v1.17.0-ce

v1.16.1-ce

What's Changed

• fix(space): replace hardcoded engine version with constant and use FindSpaceLatestVersion by @phantom-rabbit in #773

• Fix repository pagination bug by @pulltheflower in #826
• Add temporal GetSystemInfo timeout config by @pulltheflower in #826
• enable VLLM eager by default by @ganisback in #826

Full Changelog: v1.16.0-ce...v1.16.1-ce

v1.16.0-ce

v1.16.0-ce

✨ New Features

• Advanced Agent Services
  • Introduced Agent Memory Service to provide persistent context for agent interactions.
• Hardware & Compute Enhancements
  • vGPU Support: Initial support for vGPU and physical GPU resource reporting and usage calculation.
  • AMD GPU Support: Full compatibility for both Inference and Finetuning tasks on AMD hardware.
  • CUDA Version Management: Added a new CUDA-version API, allowing resource switching based on specific CUDA version requirements.
  • Enabled multi-host inference support for NVIDIA vLLM and SGLang.
• Storage & Infrastructure
  • Added PVC support for Spaces, allowing persistent storage for application environments.
  • Support for network interface configuration in "in-cluster" mode.
• Multimodal
  • Support for Text-or-Image-to-Video (TI2V), allowing text-to-video and image-to-video generation.
• Developer Experience
  • Upgraded Gradio SDK to 6.2.0 (maintaining backward compatibility with 5.1.0).
  • Space optimization: Added ability to skip build steps for Gradio, Streamlit, Nginx, and MCP Server environments.
  • Added GET/HEAD routes for Code and MCP resources within the CSGHub SDK.

🚀 Enhancements & Bug Fixes

• Space & Deployment Improvements
  • Added status filters and availability filters for the Space Index API.
  • Support for public Docker registry configuration and updated runtime images.
  • Deployment responses now include the service name for easier identification.
  • Added the ability to force-delete deployments.
• Repository & Mirroring
  • Enhanced repo syncer with retry mechanisms for mirror tasks.
  • Fixed issues where repositories total count mismatched.
  • Reset running mirror tasks automatically after a mirror LFS service restart.
  • Mirror tokens are now strictly limited to read-only actions for improved security.
• Safety & Content
  • Refactored sensitive scenario handling and updated vocabulary for better moderation accuracy.
  • Updated moderation service risk levels.
• General Fixes
  • Fixed token usage recording for streaming chat completion requests.
  • Resolved GGUF listing issues and collection message bugs.
  • Fixed specific bugs in Notebook environments and resource checking logic.
  • Improved repository sync status consistency between CE and EE versions.
• Refactoring
  • Significant architectural cleanup including splitting repository implementations from interfaces and using cluster pool interfaces.

New Contributors

• @denny-zhao made their first contribution in [#718]

Full Changelog: v1.15.1-ce...v1.16.0-ce

v1.15.1-ce

What's Changed

• Cherry pick 1.15 by @ganisback in #745

Full Changelog: v1.15.0-ce...v1.15.1-ce

• Fix ce ee sync status bug (https://github.com/OpenCSGs/csghub-server/pull/740[)](https://github.com/OpenCSGs/csghub-server/commit/e217c63ea3a899437d68b78c62822d1f1f64422d)
• Cherry pick 1.15 (https://github.com/OpenCSGs/csghub-server/pull/745[)](https://github.com/OpenCSGs/csghub-server/commit/63af6dd320695dfb10222a580af49f39040c4445)

v1.14.2-ce

Full Changelog: v1.14.1-ce...v1.14.2-ce

• Fix repositories total mismatch bug
• Fix xnet header returned if file is lfs

v1.15.0-ce

What's Changed

• build(deps): bump github.com/containerd/containerd from 1.7.27 to 1.7.29 by @dependabot[bot] in #511
• Sync allFiles changes by @pulltheflower in #563
• fix add time filtering for space log by @phantom-rabbit in #556
• Enhance repo creation by @pulltheflower in #570
• Optimize aigateway: refactor moderation component by @QinYuuuu in #543
• add recharge deposited notification by @QinYuuuu in #545
• Optimize cluster usage API performance through concurrent cluster queries by @QinYuuuu in #550
• print debug log by @QinYuuuu in #558
• add 'docker' tag to test and build ci by @QinYuuuu in #559
• add error i18n by @QinYuuuu in #569
• update i18n and ut by @QinYuuuu in #565
• Error scan cmd by @QinYuuuu in #566
• Hotfix for modelscope repo sync by @QinYuuuu in #560
• support flex gpu label type by @QinYuuuu in #555
• Reject git push in xnet enabled repo by @QinYuuuu in #564
• Add missing field for file by @pulltheflower in #577
• add pr_dependency_check by @QinYuuuu in #587
• Remove useless file and merge mcp resource by @HaiHui886 in #596
• optimize token counter for ut and fix token counter support different prompt struct by @QinYuuuu in #590
• Revert "add 'docker' tag to test and build ci" by @QinYuuuu in #578
• fix lint issue by @luojun96 in #599
• fix add instal log completa by @phantom-rabbit in #579
• Add admin repos API by @pulltheflower in #601
• Enhance logs by @QinYuuuu in #544
• Fix cluster info by @QinYuuuu in #561
• Fix xnet dataset upload bug by @QinYuuuu in #562
• remove embedding moderation by @QinYuuuu in #567
• Xnet csghub sdk by @QinYuuuu in #573
• Fix xnet refresh route bug by @QinYuuuu in #568
• Support tts by @QinYuuuu in #602
• Support function call by @QinYuuuu in #613
• Add NotebookType support and update scene validation by @QinYuuuu in #606
• fix upload databaset error by @QinYuuuu in #607
• Return empty response if repo has no files by @QinYuuuu in #612
• feat: build image name refactoring by @QinYuuuu in #616
• optimize ChatCompletionRequest JSON marshal and unmarshal，support unknown data by @QinYuuuu in #611
• fix swag fail by @QinYuuuu in #614
• add ut for repo full check workflow by @QinYuuuu in #618
• [dataviewer] Enhance UUID column type handling for messy value in dataviewer by @QinYuuuu in #620
• fix framework tag recoginization by @QinYuuuu in #605
• feat image sensitive check support url by @QinYuuuu in #588
• Add httpInsecureSkipVerify options for repo sync by @QinYuuuu in #575
• Add CreateBranch method by @QinYuuuu in #623
• fix inference name and model tag by @QinYuuuu in #604
• feat(inference): Support create, query, traffic control and delete for inference service versions by @QinYuuuu in #608
• Fix request id dup in rsp header by @QinYuuuu in #595
• Non admin user can quit organization by @QinYuuuu in #624
• Add SetDefaultBranch method by @QinYuuuu in #626
• style: embedding token counter use interface instead of struct by @QinYuuuu in #627
• add error handle for space resource create by @QinYuuuu in #628
• fix request for embedding by @QinYuuuu in #629
• add space resource list hardware types by @QinYuuuu in #630
• update user service by @luojun96 in #642
• fix accounting for replica cases by @QinYuuuu in #631
• Fix deploy mcp server with default branch master fail by @QinYuuuu in #632
• sync user invitation interface. by @luojun96 in #643
• feat: can custom docker namespace for imagebuilder by @QinYuuuu in #633
• Refactor(log): Add ctx context to all log prints in runner and user by @QinYuuuu in #625
• add filter to index space resource by @QinYuuuu in #634
• fix: PVC name in lowercase by @QinYuuuu in #635
• Merge account check function by @HaiHui886 in #650
• Merge use_limit column changes by @HaiHui886 in #654
• feat: accounting price list support query hardware type by @QinYuuuu in #636
• Add cache for model and dataset info API by @pulltheflower in #649
• Feat api limit for ip geo by @QinYuuuu in #637
• Fix search by industry tag no data bug by @QinYuuuu in #640
• Separate ce EE saas routes by @QinYuuuu in #638
• refactor user component to split phone related functions to single component. by @QinYuuuu in #641
• feat: optimize inference deployment instance logs by commit_id by @QinYuuuu in #644
• add metrics for runner webhook by @QinYuuuu in #647
• upgrade golang to 1.25.5 and use synctest package for concurrent unit tests by @QinYuuuu in #651
• Support query metering data without instance_name and update filtering logic by @QinYuuuu in #652
• Feat aigateway record usage support filter by @QinYuuuu in #655
• add finetune logs by @QinYuuuu in #656
• Update dependencies in Dockerfiles (coagent-python, huggingface_hub, gradio) by @QinYuuuu in #657
• docs: update space builder base README by @QinYuuuu in #658
• clean deploy code by @QinYuuuu in #659
• add query to list model by @QinYuuuu in #660
• update makefile by @QinYuuuu in #661
• Add git tree operation timeout config by @QinYuuuu in #662
• fix Revision delete using logical deletion by @QinYuuuu in #663
• add create table knative_service_revision in runner by @QinYuuuu in #664
• fix: update user login time on login by @QinYuuuu in #666
• update index for knative_revision by @QinYuuuu in #667

Full Changelog: v1.14.0-ce...v1.15.0-ce

v1.14.1-ce

What's Changed

• Fix wrong base image path for v1.14.0-ce by @HaiHui886 in #683
• Fix missing of mirror current task by @pulltheflower in #694

Full Changelog: v1.14.0-ce...v1.14.1-ce

v1.14.0-ce

v1.14.0-ce

✨ New Features

• Deployment Workflow Migration: Refactored the core Deploy Scheduler and successfully migrated its logic to use Temporal Workflow for improved reliability and scalability. ([#495])
• Finetuning Job Support: Added support to submit fine-tune tasks directly as standard jobs. ([#522])
• AI Gateway Integration: The AI Gateway now supports the MCP (Multi-Cluster Platform) for enhanced model serving capabilities. ([#529])
• Jupyter Notebook Support: Application space deployment now supports Jupyter Notebook environments.
• Internal Notification Tool: Introduced a new tool to generate basic internal notification messages. ([#551])

🚀 Enhancements & Bug Fixes

• Dependency Updates: Bumped argo-workflows from 3.5.13 to 3.6.12. Upgraded casdoor SDK to v1.22.0 and updated user info retrieval to use UUID. ([#481], [#534])
• API & Core: Enhanced middleware robustness. Ensured header keys are handled as case-sensitive. Enhanced concurrency in CheckRepoFiles and improved test coverage. Enabled soft delete functionality for licenses. ([#547], [#482], [#530], [#553])
• Deployment & Runner: Improved runner network discovery logic to read clusterIP if no ingress IP is found. Added StorageClass support in cluster report events. ([#484], [#525])
• Finetuning: Added model/dataset fields to finetune job operations for richer metadata. ([#527])
• Integration: Added User-Email to request headers for DataFlow and Label Studio integrations. ([#492])
• Git/LFS Fixes: Fixed a client bug preventing the deletion of remote branches. Fixed a cache bug in LFS sync upload ID. ([#521], [#489])
• Temporal Workflow Fixes: Fixed command errors and a worker panic when the sensitive checker was not enabled. Added initialization sensitive checker in the temporal worker. ([#501], [#506], [#519])
• Multi-Sync & Rproxy Fixes: Corrected the logic for setting the service host name in the rproxy header. Fixed a bug where multi-sync MCP servers failed to display file lists and READMEs. ([#485], [#533])
• General Bug Fixes: Fixed bugs related to getting evaluations with access control, AI Gateway external LLM support, and repository creation. ([#529], [#532], [#536])

Full Changelog: v1.12.0-ce...v1.14.0-ce