v0.1.0-rc.8

What's Changed

New Features

OpenRouter Support

Use any model available on OpenRouter with the new -m/--model CLI flag:

codey -m 'openrouter::anthropic/claude-3.5-sonnet'
codey -m 'openrouter::google/gemini-2.0-flash-thinking-exp:free'

Sub-Agent Architecture Overhaul

Completely redesigned sub-agent system for better performance and reliability:

• Tool approval bubbling - Sub-agent file edits now prompt for user approval with [agent-label] prefix
• Label-based management - spawn_agent takes a label, get_agent retrieves by label
• Concurrent execution - Per-agent foreground lanes for parallel sub-agent work
• Effect polling system - Non-blocking coordination for IDE previews and approvals
• Better background performance - Agents run in main event loop instead of isolated background tasks

NeoVim Multi-Instance Support

Multiple codey instances can now connect to the same NeoVim socket:

• Selection events broadcast to all connected instances
• Preview coordination with atomic claiming (one preview at a time)
• Automatic polling when preview slot is occupied

Improvements

genai 0.5.x Upgrade

Migrated from local patches to git dependency with three upstream contributions:

• Anthropic thinking blocks - Proper Thinking struct with content + signature for tool use continuation
• Streaming cache tokens - Accurate context size display (was showing ~18 tokens, now shows ~200k)
• OAuth support - Authorization header detection for Claude integrations

Bug Fixes

• Fix OpenAI tool call streaming panic when tool index > 0
• Fix sub-agent IDE preview deadlock with effect polling system
• Skip reasoning_effort option for non-Anthropic models

Full Changelog: v0.1.0-rc.7...v0.1.0-rc.8