cc-haha changelog: what's new each month
Every stable cc-haha release summarised by month: the highlights, new features, improvements, fixes and anything you need to act on. 3 months covered; the current month updates daily.
October 2026 · so far
7 releases: 0.6.8 → 0.7.5
October brought a complete visual overhaul with the new Porcelain design system, major improvements to voice transcription and workspace document handling, and powerful new features like cross-session model distribution and editable message history.
Action needed
- Trace list, trace tab, and independent trace windows removed; existing saved traces migrated to session tags.
- Regular SubAgent concurrency now limited per session (default unlimited); configure in Settings > General.
- Agent Teams interface unified to singular naming; explanatory modal shown before enabling with token usage warning.
Highlights
- Redesigned desktop and mobile interfaces with the new Porcelain visual system, featuring unified styling, streamlined navigation, and improved readability across all pages.
- Local voice transcription with microphone button in input area, supporting multiple languages with customizable settings and waveform preview.
- Edit and re-run sent messages with automatic rollback detection and confirmation when changes affect downstream rounds or files.
- New curated skills directory with 398 pre-selected skills across 13 categories, security reports, and safety confirmations for unaudited skills.
- Cross-session model distribution allowing a conversation to delegate tasks to sub-conversations, each with independently chosen models and providers.
New
- Local voice transcription with audio resampling and language selection, with results appearing in draft before sending.
- Workspace document preview for PDF, Word, Excel files opening directly in the sidebar with zoom, scrolling memory, and auto-refresh.
- Conversation Trace view showing high-density event log with system prompts, user messages, context injection, assistant replies, and tool calls.
- Safe data directory migration with staging, validation, and recovery logs to prevent data loss during moves.
- ChatGPT OAuth Fast mode for faster inference with variable speed depending on model and potential additional token consumption.
- CSV and TSV table preview in workspace with source code toggle and support for multiple character encodings.
- Session tab limit control (5/8/12/20 or unlimited) with option to restore closed tabs and preserve important sessions.
- DeepSeek account login without API key, auto-migrating existing model mappings.
Improved
- Input menu reorganized into task-specific (skills/connectors/files) and run modes (Teams/computer control) with clearer connector status display.
- Voice transcription recording bar shows waveform, countdown, and send option; default enabled on fresh installs.
- Long task streaming output no longer blocks window operations like creating sessions or switching tabs.
- Context compression now shows real-time progress with elapsed time and generated character count.
- Long sessions with round modifications load preview much faster without repeated file scanning.
- Terminal panel functional on Linux packages with proper native module compilation and glibc 2.28 compatibility.
- Input field overlay (slash commands, @mentions, +menu) respects available space above and scrolls internally instead of being cut off.
- Upstream API errors occurring in SSE streams now use full retry budget instead of single retry.
Fixed
- Windows desktop repeatedly showing black terminal windows when opening app or creating/restoring sessions.
- Bash commands randomly killed due to process guardian pipe conflicting with Bun file descriptor reuse.
- Text selection menu across sessions applying to wrong conversation when identical messages exist.
- Compression repeatedly retrying after hitting context limits instead of stopping at three consecutive failures.
- macOS full-screen mode now uses native system full-screen with independent Space instead of just enlarging window.
- Missing prompt token usage in cache hit rate calculation when tokens were already counted in upstream response.
September 2026
8 releases: 0.6.0 → 0.6.7
September brought major improvements to computer use on macOS and Windows, expanded messaging platform support, and significant UI refinements including cross-session collaboration and smarter context management.
Action needed
- Removed per-app approval prompts; replaced by single Computer Use authorization when enabling the feature.
- Moved from per-app restrictions to global Computer Use authorization with per-app target validation.
- Agent page model aliases (e.g., opus) now resolve through vendor mapping; third-party vendors map to their own model slots.
Highlights
- Native macOS engine with independent virtual cursor and more reliable background window capture, plus improved Windows input handling.
- New messaging adapters for Feistel, WeCom, QQ, and Slack with scan-to-bind setup and persistent connections.
- Cross-session chat and control: Claude can now create and manage multiple parallel work sessions, reference previous conversations, and coordinate between them.
- Persistent JavaScript execution for Computer Use batches known actions without model round-trips, and improved accessibility-based clicking.
- Sidebar chat feature for quick questions without interrupting main tasks, plus Agent Teams pre-launch review and auto-answer for waiting questions.
New
- Persistent isolated JavaScript worker execution for Computer Use to batch actions without model round-trips.
- Feishu bot creation via QR code scanning in desktop settings with automatic credential save.
- WeCom, QQ, and Slack adapters with scan-to-bind setup and persistent socket connections.
- Cross-session message and control tools: ListSessions, ReadSession, CreateSession, SendSessionMessage, WaitSessions.
- Session referencing in chats with @ mentions to search and link previous conversations without copying full history.
- Sidebar chat mode (/btw) for temporary parallel conversations inheriting parent session context.
- Skills and Connectors unified management page with install, disable, and remove options.
- OpenCode Go subscription provider supporting GLM, Kimi, MiniMax, Qwen, Grok, GPT models with automatic protocol routing.
Improved
- macOS gesture, keyboard, app list, screenshot, scroll, and clipboard behavior alignment with regression test coverage.
- Coordinate clicking now uses precise accessibility hit detection instead of pixel-based clicking.
- Screenshot includes pixel dimensions while sending limited resolution to model, preserving evidence without context bloat.
- Cancel operations now work reliably across background processes and lock acquisitions.
- Session history retention defaults to 365 days (configurable in Settings with cleanup preview).
- Tab management with drag-to-reorder, batch close, and reopen closed tabs; terminal movable between sidebar and bottom.
- File tree single-click to preview and double-click to pin; output cards fully clickable; improved Chinese filename handling.
- Context usage panel redesigned to show remaining window as primary metric with speed, cache hits, and cost displayed together.
Fixed
- Click animation resources now resolved from running native helper directory, preventing asset loading failures in production builds.
- Missing or unreadable animation frames preserve programmatic click feedback without affecting subsequent operations.
- Persistent call rewrites now preserve original statement boundaries to avoid auto-semicolon insertion changing semantics.
- Session list loading no longer falls back to full JSONL scan when local index is empty or building.
- Window drag functionality restored for sidebar and workspace top bar blank areas.
- Model-unspecified team members now follow the main session's current model instead of reverting to hardcoded Opus.
August 2026
4 releases: 0.5.2 → 0.5.5
August brought major expansions to cc-haha's multi-agent capabilities and image handling. The month focused on team coordination features, improved streaming performance, and stability across desktop and chat interfaces.
Highlights
- Agent Teams workbench now visualizes multi-agent coordination with real-time task tracking and dependency visualization.
- Dynamic Workflow orchestration lets models write and run scripts to drive subagents concurrently or in sequence.
- Image generation and editing work directly in chat conversations after signing in with supported providers.
- Generated files now open with native applications, and user-attached images can be edited in place.
- Undo for real turns restored after being broken in v0.5.3; tool activity is now folded into compact summaries.
New
- Agent Teams workbench for visualizing multi-agent teams with members, tasks, communication feed, and dependency canvas.
- Dynamic Workflow orchestration with models writing scripts to drive subagents.
- Image generation and editing built into chat conversations.
- Create git branches directly from the launch picker.
- Tool Search made opt-in per provider.
- User-attached images can be edited as ImageEdit inputs.
- Generated files open with ranked native applications menu.
- Model-aware default reasoning effort that follows the selected model.
Improved
- Model selector now supports search for faster model discovery.
- Batch element selection is now supported.
- Conversation navigation streamlined.
- Grok reasoning summaries now stream incrementally.
- First-time users without a configured model are guided through setup.
- Inline @-mentions render as pills in the composer.
- Thinking blocks in history are separated and duplicate blocks are collapsed.
- Chat image rendering stabilized with improved single and multi-image layouts.
Fixed
- Unstable GPT streaming and agent recovery issues resolved.
- Completed turns now refresh their changes properly.
- Rewind no longer loses partial snapshot changes.
- Stale activity tasks no longer linger after a turn ends.
- IM channels now list all accessible projects instead of only the default.
- Truncated tool calls no longer hang sessions; broken fragments are terminated safely.
Summaries are written automatically from the official release notes (full changelog ↗); check the original notes before relying on a detail. cc-haha: pricing, features and alternatives · All changelogs