docker/docker-agent v1.36.0

on GitHub

This release adds WebSocket transport support for OpenAI streaming, introduces configurable tool call token limits, and improves the command-line interface with new session management capabilities.

What's New

• Adds WebSocket transport option for OpenAI Responses API streaming as an alternative to SSE
• Adds /clear command to reset current tab with a new session
• Adds configurable max_old_tool_call_tokens setting in agent YAML to control historical tool call content retention

Improvements

• Hides agent name header when stdout is not a TTY for cleaner piped output
• Sorts all slash commands by label and hides /q alias from dialogs, showing only /exit and /quit
• Injects lastResponseID as previous_response_id in WebSocket requests for better continuity

Bug Fixes

• Fixes data race on WebSocket pool lazy initialization
• Fixes panic in WebSocket handling

Technical Changes

• Removes legacy syncMessagesColumn and messages JSON column from database schema
• Simplifies WebSocket pool code structure
• Documents external OCI registry agents usage as sub-agents

What's Changed

• Add WebSocket transport for OpenAI Responses API streaming by @dgageot in #2186
• feat: make maxOldToolCallTokens configurable in agent YAML by @gtardif in #2192
• Remove syncMessagesColumn and legacy messages JSON column by @dgageot in #2196
• docs: document external OCI registry agents as sub-agents by @dgageot in #2195
• Support echo "hello" | docker agent | cat by @dgageot in #2197
• Hide /q from dialogs and sort all commands by label by @dgageot in #2200
• Add /clear command to reset current tab with a new session by @dgageot in #2199

Full Changelog: v1.35.0...v1.36.0