29 posts tagged with "platform"

Release description not available.

🚀 Highlights​

• Kai updated to leverage the latest Aleph APIs for faster and more efficient responses

🛠 Improvements​

• Kai system prompt optimization
  • Updated to take advantage of the latest Aleph APIs.
  • Improved efficiency and response speed.

🚀 Highlights​

• Patch release improving cloud job reliability

🐛 Fixes​

• Fixed a race condition that could prevent some cloud jobs from starting correctly.

🚀 Highlights​

• Fix for a Kai error affecting certain prompts
• Runtime upgrade for improved performance and security

🐛 Fixes​

• Fixed an issue in Kai that could trigger a "finishReason" error for certain prompts.

🛠 Improvements​

• Runtime upgrade
  • Improved overall performance.
  • Enhanced security and stability.

🚀 Highlights​

• Patch release addressing Kai credit usage issues

🐛 Fixes​

• Updated Kai configuration to enforce a specific AI provider, resolving credit-related issues experienced by some users.

🚀 Highlights​

• Improved handling and visibility of transient errors
• Internal groundwork for upcoming bindings

🛠 Improvements​

• Transient error handling

  • More robust detection and reporting of temporary failures to improve reliability and troubleshooting.

• Platform readiness

  • Background updates preparing the platform for upcoming bindings and integrations.

🚀 Highlights​

• Improved API documentation and SDKs
• runtimeId is now optional when creating executions and jobs

🛠 Improvements​

• API & SDK enhancements

  • Clearer API documentation.
  • Updated SDKs for better developer experience and consistency.

• Simplified execution & job creation

  • runtimeId is now optional when creating executions or jobs.
  • Defaults automatically to the current Aleph version if not specified.

🧪 Migration Notes​

• Existing integrations can omit runtimeId unless a specific runtime version is required.

🚀 Highlights​

• Internal pre-release focused on improving interoperability and integration scenarios
• No user-facing changes in this release.

🛠 Improvements​

• Backend enhancements to support broader integration use cases.
• Internal changes preparing the platform for upcoming interoperability features.

🚀 Highlights​

• Kai model updated to improve AI code generation quality

🛠 Improvements​

• Kai model upgrade
  • Improved code generation accuracy and consistency.
  • Better adherence to project structure and conventions.
  • More reliable diffs and suggested changes.

🚀 Highlights​

• Improved execution diagnostics with signal capture
• Safeguard added to prevent excessive file uploads during executions

✨ New Features​

• Execution signal capture
  • In addition to exit codes, the platform now captures the process termination signal.
  • Provides clearer insight into whether a process exited normally or was interrupted (e.g., killed, timeout, etc.).

🛠 Improvements​

• Upload safety guard
  • If more than 100 files are created or modified during an execution:
    • All affected files are bundled into a single ZIP archive.
    • The archive is uploaded as-is to the platform.
  • Prevents rogue or misconfigured scripts from uploading excessive numbers of files.

🧪 Migration Notes​

• No action required for users.
• Workflows generating large numbers of files may now produce a ZIP artifact instead of individual files.

🚀 Highlights​

• Patch release fixing file listing pagination

🐛 Fixes​

• Fixed an issue affecting pagination when listing files.

🚀 Highlights​

• Kai updated to better support figures and admonitions in generated content

🛠 Improvements​

• Kai system prompt update
  • Improved support for generating figures.
  • Improved handling of Markdown admonitions (note, tip, warning, etc.).

🐛 Fixes​

• Minor internal improvements to AI content formatting.

🚀 Highlights​

• Internal platform updates and maintenance release

🚀 Highlights​

• Patch release fixing default timeout handling for Cloud Agents

🐛 Fixes​

• Fixed handling of default timeouts for Cloud Agents to ensure executions behave as expected.

🚀 Highlights​

• Cloud agents now report accurate system information
• Improved error reporting for cloud job scheduling failures

🛠 Improvements​

• Accurate cloud agent system info

  • Cloud agents now correctly report CPU core count and total memory.

• Better cloud job error reporting

  • Jobs that fail due to cloud provisioning issues now surface clear failure states instead of hanging in scheduled or staled status.

🐛 Fixes​

• Minor fixes to improve cloud execution reliability.

🚀 Highlights​

• CPU usage metrics updated to follow Linux conventions
• New Cloud Agents for Jobs offering dedicated, high-performance machines (limited preview)
• Minor bug fixes and performance improvements

✨ New Features​

• Cloud agents for jobs
  • Jobs can now run on dedicated cloud agents with powerful machine configurations.
  • Designed for heavy, long-running, or performance-sensitive workloads.
  • Limited preview for now, we will open it up gradually

🛠 Improvements​

• CPU percent metrics aligned with Linux standards
  • CPU usage now reports 100% = one CPU core, instead of a percentage of total machine CPUs.
  • Metrics are now consistent with standard Linux tools and expectations.

🐛 Fixes​

• Minor bug fixes and general performance improvements.

⚠️ Breaking Changes​

• CPU metrics interpretation has changed
  • CPU percentages now follow Linux semantics.
  • Jobs executed before this release will not be retroactively updated and may show inconsistent CPU values.

🚀 Highlights​

• Point release focused on security and stability improvements

🛠 Improvements​

• Security hardening

  • Tighter API security headers and CORS handling to better protect requests.

• Stability improvements

  • Improved backend reliability for smoother execution and job handling.

🐛 Fixes​

• Minor fixes to improve overall robustness.

⚙️ Internal / Technical Changes​

• Minor update to the Agent Cloud Model to simplify alignement

🚀 Highlights​

• Renamed Communal Agent to Interactive (Shared) for clarity
• Improved load balancing for shared executions
• Added in-product guidance for understanding cloud agents

✨ New Features​

• Interactive (Shared) agent

  • The former Communal Agent is now called Interactive (Shared).
  • Provides a shared, cloud-hosted execution environment for interactive workloads.

• Cloud agent tooltip

  • Added a tooltip to explain the cloud agent model and when to use shared vs personal agents.

🛠 Improvements​

• Better load balancing
  • Improved execution distribution on shared agents for more consistent performance.

⚙️ Internal / Technical Changes​

• Refactored communal agents into a more general Cloud agent concept.

🚀 Highlights​

• Communal agents are now available for direct executions (internal user only ATM).

✨ New Features​

• Communal agents
  • A shared, load-balanced agent is now available for direct executions.
  • Communal agents are discoverable via the agents list for users who have access to it.
  • Designed for lightweight, shared execution without requiring an agent.

🚀 Highlights​

• AI Assistant now produces shorter, clearer chat summaries
• Improved diff generation with stricter rules for cleaner, more unified patches
• AI Assistant can now create Markdown files in addition to Aleph code

✨ New Features​

• Markdown file creation via AI Assistant
  The AI Assistant can now generate Markdown files directly, alongside Aleph source files.

🛠 Improvements​

• Shorter AI chat summaries
  Chat summaries are now more concise and focused, making conversations easier to scan and review.

• Stricter unified diff generation

  • Tighter rules ensure diffs are more consistent and easier to apply.
  • Reduces malformed or overly fragmented patches.
  • Improves reliability when applying AI-generated changes.

🚀 Highlights​

• System-level metrics are now collected during executions for better observability
• AI Chat assistant has been renamed
• AI Chat system prompt updated to significantly reduce hallucinations and improve API correctness

✨ New Features​

• Execution system metrics collection
  • The agent now periodically collects:
    • CPU usage
    • Memory usage (Resident Set Size)
  • For direct executions, a summarized metrics report is sent at the end of the execution.
  • For job executions, all collected samples are streamed to the API periodically during runtime.

🛠 Improvements​

• Improved AI Chat reliability
  • Updated system prompt to enforce stronger adherence to Aleph APIs.
  • Reduced hallucinations and off-spec responses when generating code or explanations.

🚀 Highlights​

• Major refactor of the AI Chats API to improve reliability, extensibility, and alignment with the OpenAI API
• Streaming AI responses now supported for real-time content generation
• Backend now supports feature flags for safer rollouts
• New APIs added to update job names and notes

✨ New Features​

• Streaming AI responses
  AI-generated content can now be streamed as it is produced, enabling more responsive user experiences.

• Client + server tool support
  The AI API now supports client-side tools in addition to server-side tools, unlocking more flexible AI workflows.

• Job metadata API
  New endpoints allow updating job names and notes

• Backend feature flags
  Feature flag support added in the platform (in addition to the Workshop) to progressively enable or disable features without redeploying.

🛠 Improvements​

• AI API aligned with OpenAI’s canonical format
  The AI Chats API now uses a messages/roles structure consistent with the OpenAI API, simplifying integrations and future compatibility.

• Improved editor context handling
  Editor context is now explicitly passed and correctly processed by the model.

• Usage tracking per conversation

🐛 Fixes​

• Streaming endpoints now return errors correctly instead of silently failing.
• Improved request lifecycle handling to prevent orphaned or hanging AI requests.

⚙️ Internal / Technical Changes​

• Refactored AI tools container injection into a dedicated module.
• Added graceful cancellation of in-flight AI requests when the API shuts down (via AsyncDisposable callbacks).
• Client package updated to match the new AI API behavior and contracts.

⚠️ Breaking Changes​

• AI Chats API has been significantly refactored
  • Request and response formats have changed
  • System prompt behavior is different
  • Streaming behavior is now first-class

⚠️ This will break the existing AI Chat feature in production until the Workshop is updated.

🚀 Highlights​

A clearer and more capable AI experience behind the scenes, with improved model flexibility and stronger Aleph code support.

⚙️ Internal / Technical Changes​

• Internal users can now select which AI model to use.
• AI Chats now include Aleph code parsing and validation during code generation.

🚀 Highlights​

• Introduction of API tokens to support secure automation and machine-to-machine integrations.
• Improved internal AI Chat tooling and a more secure SSE/EventSource implementation in the TypeScript client.

✨ New Features​

• API Tokens — Enable automated workflows and machine-to-machine integrations. Each token is tied to a workspace and carries its own identity and permission scope.

⚙️ Internal / Technical Changes​

• Internal users can now choose between multiple presets for AI Chat features.
• The TypeScript client package now uses a refactored SSE/EventSource implementation with improved security.

🚀 Highlights​

• Improved onboarding flow for new users.
• Early groundwork completed for upcoming AI-assistant features.

✨ New Features​

• Personal workspace initialization — New users now receive a pre-populated workspace on their first login, making it easier to get started immediately.

🐛 Fixes​

• Resolved an intermittent issue where rate limits on job and user listings could cause partial results. A caching layer now reduces request pressure and prevents these failures.

⚙️ Internal / Technical Changes​

• Pre-release of the AI Chats feature is now available internally for testing and validation.

🚀 Highlights​

• Fix a bug that prevented some users to search in files
• Update agent reconnection logic

🛠 Improvements​

• Agents now wait more predictably between connection instances, and give up after 3 days of unsuccessful attempts
• Agents logging is less verbose

🐛 Fixes​

• Resolve a bug in search that prevented users to search in files if one of the file is empty

🧪 Migration Notes​

• Agents will self-update automatically

⚙️ Internal / Technical Changes

• Add internal endpoints for management

🧩 Fix

Commented include statements in Aleph scripts are no longer parsed as dependencies by the Agent. Example:

// include "dependency.aleph"

Previously, this line could incorrectly trigger dependency resolution and fail the execution. It is now correctly ignored.

🔧 Impact

This fix prevents incorrect dependency lookups and ensures consistent execution graphs when using commented or disabled includes.