0.125.0

• Refresh auth token before each LLM API call, preventing stale tokens during long-running tool calls.
• Add background shell command support via background parameter on shell_command tool and new bg_job tool for managing long-running processes. #77
• Add disabled field to MCP server notifications, allowing clients to distinguish between stopped and config-disabled servers. #403

0.124.5

• Fix Github Enterprise models fetch. #402

0.124.4

• Fix github-copilot retry loops in chats.
• Fix Github Enterprise to use the proper url during prompts. #402

0.124.3

• Fix /resume broken for OpenAI chats: handle nil reasoning text during replay, preserve prompt-id after chat replacement, and clear UI before replaying messages. #400
• Add GitHub Enterprise support for Copilot authentication via auth.url and auth.clientId provider config. #402
• Add chat/update notification for renaming chats. Chat titles are now persisted to the database and broadcast to all connected clients including remote web interface.

0.124.2

• Fix OpenAI Responses API tool calls not executing when streaming response returns empty output, and fix spurious retries caused by stale tool-call state with Copilot encrypted IDs. #398

0.124.1

• Add cacheRetention provider config for Anthropic to support 1-hour prompt cache TTL. Set to "long" for sessions with pauses longer than 5 minutes.

0.124.0

• Add chatRetentionDays config to control chat and cache cleanup retention period, default changed from 7 to 14 days. Set to 0 to disable cleanup. #393
• Preserve full chat history across compactions using tombstone markers instead of replacing messages. #394
• Add message flags — named checkpoints for resuming and forking chats. #395
• Fix OpenAI models getting stuck at toolCallPrepare when streaming response returns empty output in response.completed. #398

0.123.3

• Fix exceptions on openai responses models when creating tasks.
• Fix potential infinite auto-compact loop when context overflow persists after compaction. #391
• Improve Anthropic prompt caching: split system prompt into static/dynamic blocks, add cache markers to the tools array, and memoize static instructions per chat.

0.123.2

• Wait for pending MCP tool list refresh before reading tools after tool execution, fixing race where dynamically loaded tools were not immediately available.

0.123.1

• Fix OAuth HTTPS server crash in native image by building SSLContext in-memory instead of relying on ring-jetty's keystore path reflection.