v0.12.15

hotfix

v0.12.14

What's Changed

🐛 Bug Fixes

• opencode: Fixed issue preventing Kimi model usage due to missing reason_content field
• custom_llm: Improved tool parsing for custom LLM services
• setup: Added openai_codex to the memory LLM configuration

v0.12.13

🚀 Release Notes

✨ New Features

• Provider: Introduced OpenCode Go as a new provider

🐛 Bug Fixes

• LLM: Corrected parameters for Qwen models

🧹 Maintenance & Chores

• LLM: Removed deprecated model
• Docker: Pinned the a2a-sdk version for build stability

---

📝 Summary

This release adds OpenCode Go as a new provider option, fixes Qwen model parameter configuration, removes a deprecated LLM model, and pins the a2a-sdk Docker dependency for more predictable builds.

v0.12.12

✨ Features

• LLM: Added Qwen 3.6 support (8a8ae0d)
• Agent: Allowed agents in remoting mode to have memory capabilities (9ab9ab4)

🔧 Improvements

Voice

• Made voice system more robust (eb982e7)
• Refined all voice-related code (25ce1a1)
• Clean audio buffer if voice is not completed (68720b0)
• Corrected voice behavior (c1f8eb2)
• Made voice more natural (ca005ad)
• Added more speak usage scenarios (b2b3e7b)
• Made agent speak more verbose (304c2b3)
• Speak tool now only available when user sets it via argument (63fca05)

Memory

• Fixed bug where search memory could miss entries in to_date range (228362a)
• Allowed passing session_id to maintain agent across a2a with clean memory (d9e3fab)
• Decoupled memory tool and agent manager; now uses registrar as source of agent (9e3b6b7)

General

• Removed deprecated function (29a1906)

v0.12.11

✨ Features

• A2A Memory Support: Enabled user-controlled memory on A2A workflows when explicitly setting memory-path parameter
• Environment Configuration: Added support for using environment variables to configure settings
• Voice Communication: Reworked voice functionality to enable agents to use it as a communication tool
• LLM Models:
  • Added Claude Opus 4.7 model for Copilot integration
  • Added GoldenEye model for GitHub integration

🔧 Improvements

• LLM Content: Corrected thinking content generation for improved response quality
• System Prompt: Cleaned and optimized evolved system prompt for better efficiency and clarity

v0.12.10

Release Notes

✨ Features

• Docker: Predownload model in Docker for faster inference time
• LLM: Add Gemma 4 and GLM 5.1 support in DeepInfra

🐛 Bug Fixes

• GUI: Prevent double plan display on GUI
• Fork: Fixed bug where fork title displayed the same as the main conversation

🔧 Chores & Improvements

• GUI: Group state for easier maintenance
• Conversation: Ensure conversation saves successfully if stream is interrupted

v0.12.9

🎉 Release Changelog

✨ Features

• grep: Allow grepping in multiple paths
• code_analysis: Agent can now focus on a specific scope when using read_repo

🐛 Bug Fixes

• memory: Fixed an issue that would cause crash on macOS

🔧 Chore

• general: Fixed small issue with last provider with Codex

v0.12.8

Release Notes

✨ Features

• Memory: Allow memory to collect all assistant messages during the work, not just the last message
• UI: Show the agent evaluation as the plan for transparency

🔧 Chores & Maintenance

• Transfer: Only transferable post action should be defer
• Test: Fixed test
• Browser: No need to re-compute the element UUID after action anymore
• Evolve: Refactored code of evolution
• Evolve: Correctness the evolution agent
• Log: Cleanup the log

v0.12.7

What's Changed

🐛 Bug Fixes

• Stream Module: Increased first chunk timeout and fixed JSON dump serialization
• OpenAI Integration: Ensured refresh token timing is synchronized with Codex CLI

📦 Commits

• 6c16452 - fix(stream): increase first chunk timeout and fix json dump
• 470442a - fix(openai): make sure the refresh time matched with codex cli

---

Full Changelog: View on GitHub

v0.12.6

Release Notes

🐛 Bug Fixes

• Agent Transfer: Fixed issue where transfer operations still contained planning rules that should have been removed

✨ Features

• Stream Control: Users can now stop streams immediately without waiting for timeout to occur

🔧 Chores & Improvements

• Console: Enhanced console input handling to support multiline input for custom answers and tool denial responses
• OpenAI Integration: Added support for "xhigh" reasoning effort level
• LLM Defaults: Ensured default reasoning is consistently applied across all LLM operations