v0.9.0

• Live method replacement when inspecting a running process: replace a live function in the target application with a REPL-defined function (#493).
  • #replace <method> with <replFunction> replaces the original method
  • #wrap <method> with <replFunction> wraps the original method (the replacement's first parameter is an orig delegate that calls the original),
  • #patches lists active patches
  • #revert <id>/#revert all undoes them. Instance methods take the instance as their first parameter; patches take effect immediately and persist in the target until reverted or the process exits.
  • Generic methods and pointer parameters are not supported, as well as call sites the JIT already inlined.
• Typing a built-in command (exit, clear, help) in full now submits it immediately, instead of requiring a second Enter to first accept the completion-menu item and then submit (#495).
• Breaking: the AI code-completion feature (Ctrl+Alt+Space) is now provider-agnostic. The OpenAI-specific options (--openAIApiKey, --openAIPrompt, --openAIModel, --openAIHistoryCount) have been removed and replaced with provider-neutral options: --aiApiKey, --aiPrompt, --aiModel, and --aiHistoryCount (#494).
  • --aiProvider selects a built-in preset (openai (default), anthropic, grok, deepseek, gemini, mistral, codestral), which includes a default endpoint, model, and API-key environment variable (OPENAI_API_KEY, ANTHROPIC_API_KEY, XAI_API_KEY, DEEPSEEK_API_KEY, GEMINI_API_KEY, MISTRAL_API_KEY, CODESTRAL_API_KEY). Point --aiEndpoint at any other OpenAI-compatible API to use a provider that isn't listed.
  • The default OpenAI model is now gpt-5.4-mini (was gpt-4o).