v0.5.0

Pantheon v0.5.0

Highlights

Workspace Isolation Mode — Each chat session can now optionally use its own private workspace directory, isolating files, code execution, and shell commands from other sessions.

New Features

• Two-mode workspace system (project / isolated): Default is project (shared), users can toggle to isolated mode where each chat gets a private directory at .pantheon/workspaces/{session_id}/ (#39, @1-CellBio)
• Runtime workspace mode switching via new set_chat_workspace_mode() tool — switch between project and isolated mode without creating a new chat
• All ToolSets are workspace-aware: file_manager, file_transfer, shell, notebook, python_interpreter all respect the active workspace mode
• Store CLI: Publish, install, search, and manage agent/team/skill packages (pantheon store ...)
• Store seed system: Batch import packages from factory templates and external repos (LabClaw, OpenClaw Medical, Claude Scientific, ClawBio, OmicClaw)
• PANTHEON_HUB_URL env var support for Store API client
• Evolution system improvements: Better evaluator format, real-time visualization, improved error reporting
• CLI -i flag for initial input
• Activity-aware idle cleanup via _ping responses
• Custom endpoints: Support for custom_anthropic and custom_openai providers

Bug Fixes

• Fix /keys command ProviderMenuEntry unpacking error
• Fix move_file() crash by creating parent directories before move
• Fix empty file read_file crash (start_line out of range for 0-line files)
• Fix ws://127.0.0.1 → ws://localhost for Windows IPv6 compatibility
• Fix LaTeX compilation to resolve relative figure paths
• Fix shell env over max_size issue
• Fix endpoint startup timeout (3s → 30s)

Desktop App (PyInstaller)

• Switch to onedir mode with exclude_binaries=True for correct macOS bundling
• Fix nats-server binary missing from bundle
• Fix Windows DLL corruption and read-only install location issues
• Handle fakeredis/lupa compatibility

Contributors

• @Nanguage
• @1-CellBio — session workspace isolation (#39)
• @zqbake

v0.4.10

v0.4.10: Pantheon Store, Custom Endpoints, Docker & Bug Fixes

Pantheon Store

• Store CLI (pantheon store) for publishing, installing, searching, and managing skill/agent/team packages
• Store seed system for batch-importing packages from external repos (LabClaw, OmicClaw, ClawBio, Claude Scientific)
• Install state tracking with local manifest (~/.pantheon/store_installs.json)
• RPC methods (install_store_package, get_installed_store_packages) for frontend Store dialog integration
• Skill file bundling with SKILL.md directory format and source attribution

Custom Endpoints

• Flexible custom_models configuration replacing the old unified LLM_API_BASE/LLM_API_KEY proxy
• Support CUSTOM_ANTHROPIC_* and CUSTOM_OPENAI_* environment variables
• Interactive /keys command for custom endpoint management
• Custom model aliases with highest priority in provider detection

Docker

• Dual-mode Dockerfile with GitHub Actions CI/CD workflow
• Docker entrypoint supporting both CLI and UI modes

CLI

• -i flag for initial input: pantheon cli -i "analyze my data" starts REPL with an immediate command

Infrastructure

• Activity-aware idle cleanup — _ping responses now report active threads/tasks so Hub can skip cleanup for busy pods
• Shell env fix for oversized environment variables in package runtime
• Setup wizard SKIP_SETUP_WIZARD env var support

Bug Fixes

• LaTeX PDF compilation — compile in source directory so relative figure paths (../figures.png) resolve correctly
• Speech-to-text — handle base64-encoded audio over JSON transport; add transcription timeout and spinner UI
• Context menu z-index — teleport WorkspacePanel context menu to body so it renders above chat cards
• Remove unused FilesystemTab.vue legacy component
• Skip Hub API call in local mode to avoid connection errors
• Fix test failures and Docker Hub repo name

v0.4.9

v0.4.9: CLI feedback flow, think tool display, structured questions

CLI Improvements

• Unified Review Dialog with file tabs, structured questions, and text input
• Provide Feedback flow for plan rejection — feedback is sent directly to the agent
• Think tool displays thought preview with 💭 icon instead of just showing "think"
• Notification messages rendered as Markdown (Rich library)
• Backend version exposed in _ping RPC response

Task Workflow

• Structured questions in notify_user (single/multiple choice, text input)
• Questions auto-interrupt to wait for user answers

Agent & Team

• Improved team delegation prompt
• FastMCP 3.x namespace API compatibility
• New agentic general prompt template

Bug Fixes

• Show "✓ Approved" instead of confusing "Submitted 0 answer(s)" for file-only reviews

v0.4.8

What's New

Background Task Management UI

• Add background tasks section in Timeline sidebar with collapsible list/detail views
• Show running/completed task counts with animated indicators
• Support cancel and remove actions on tasks
• Auto-refresh via NATS streaming events
• Click-to-navigate from tool calls to background task details
• Compact notification line in timeline when tasks complete
• Show spinning indicator on chat sessions with running background tasks

CLI Improvements

• Show compact colored notification in REPL when background tasks complete
• Hide bg task notification messages from chat history replay

Bug Fixes

• Fix NATS payload overflow when loading chats with large tool responses (drop oversized raw_content >50KB)
• Fix FastMCP mount() compatibility for both 2.x (prefix) and 3.x (namespace)
• Wrap bg task auto-chat notification in tags so UI can filter it from chat display

v0.4.7

What's New

Model Updates

• Add GPT-5.4 and GPT-5.4-pro as default OpenAI models
  • high: gpt-5.4-pro → gpt-5.4 → gpt-5.2-pro → gpt-5.2
  • normal: gpt-5.4 → gpt-5.2-codex → gpt-5.2 → gpt-5
  • low: gpt-5.3-chat-latest → gpt-5-mini → gpt-5-nano → gpt-4.1-mini

Misc

• Update copyright year to 2025-2026

v0.4.6

What's New

Kimi Coding Support

• Add User-Agent header for kimi-for-coding model
• Add Kimi (Moonshot) to documentation

Model Persistence Fix

• Fix /model command not persisting across restarts
• Case-insensitive agent name matching for template write-back
• Fallback source_path lookup from template manager

Smart Template Sync

• Replace force-overwrite with hash-based sync that preserves user modifications
• User changes (e.g. model set via /model) are no longer overwritten on startup

API Key Priority

• When LLM_API_BASE is set (unified proxy mode), LLM_API_KEY now takes priority over provider-specific keys

Key Management

• pantheon setup: prefix with d to delete keys (e.g. d0, d1,d3)
• /keys rm <number|name>: remove provider keys or custom endpoint from REPL

Deprecation Fix

• Use namespace instead of deprecated prefix in MCP mount

v0.4.5

What's New

Universal API Proxy Support

• LLM_API_BASE / LLM_API_KEY: New universal environment variables to route all LLM calls through a third-party proxy or custom OpenAI-compatible endpoint
• When LLM_API_BASE is active, LLM_API_KEY is automatically preferred over provider-specific keys
• Provider-specific variables (e.g. OPENAI_API_BASE) still take priority when set

Setup Wizard Enhancements

• pantheon setup: New CLI command to launch the setup wizard at any time
• Custom API Endpoint [0]: New option in the setup wizard to configure LLM_API_BASE and LLM_API_KEY interactively
• /keys 0 <base_url> <api_key>: Configure custom endpoint directly from the REPL

Other Changes

• Updated default fallback model to openai/gpt-5.2
• Documentation updated with custom API endpoint instructions

v0.4.4

What's Changed

• Fix: Include nats-ws.conf in package distribution so pantheon ui --auto-start-nats works in pip-installed environments
• Redesign: Polished README with shield badges, numbered sections, and updated branding