v1.0.0

[v1.0.0] - 2026-10-18

Added

• Added per-model reasoning-effort selection in the provider editor, covering none, minimal, low, medium, high, xhigh, max, ultra, and persistent.
• Applied configured reasoning levels to the managed Codex model catalog and /v1/models capability metadata so new models can be configured without code changes.

Changed

• Preserved custom reasoning selections when refreshing provider models and retained automatic defaults for existing configurations.
• Added DeepSeek maximum reasoning support across Responses, Chat, and Anthropic Messages, including the desktop ultra alias and upstream effort normalization.

Fixed

• Fixed local authentication configuration mismatches after upgrading from older CodexSwitch versions by synchronizing the managed provider's Bearer token independently of ChatGPT login.
• Preserved existing local inbound keys, provider credentials, and client login settings during the authentication upgrade.
• Clarified upgrade-related 401 errors and recovery instructions in Chinese, English, and Japanese.

v0.0.9

[v0.0.9] - 2026-10-01

Added

• Added a security and reliability baseline with Bearer authentication for local /v1/* APIs, cross-platform secret storage, encrypted configuration transfer, redacted diagnostics, and integrity-checked backups.
• Added OpenAI-compatible /v1/chat/completions support and model capability metadata through /v1/models.
• Added provider health checks, request correlation and upstream timing details, usage-log retention, and JSON/CSV export.

Changed

• Added configurable upstream retry handling for Responses, Chat, Anthropic Messages, HTTP streaming, and Responses WebSocket fallback flows.
• Added GPT-6.1 Sol, GPT-6 Sol, and GPT-6 Luna model capability, reasoning, pricing, and long-context support across provider catalogs, Codex synchronization, usage calculation, and logs.

Fixed

• Prevented local configuration and backup files from retaining API keys, OAuth tokens, or credentials embedded in proxy URLs.
• Preserved the previous installer after download failures and reject updates without a valid SHA-256 checksum.
• Added configuration import rollback when applying an imported configuration fails.

v0.0.8

[v0.0.8] - 2026-09-30

Added

• Added GPT-6.1 Sol to built-in OpenAI, AIOSS, and RoutinAI provider model catalogs, with a one-time upgrade that preserves existing routes and default-model selections.
• Added GPT-6.1 Sol short- and long-context pricing and Standard, Fast, Flex, and Batch rate rules. Batch pricing does not submit batch jobs.
• Added GPT-6.1 Sol max reasoning support to the managed Codex model catalog, preserving the separate ordinary and 1M context modes.

Changed

• Made GPT-6.1 Sol model prices and usage estimates service-tier-aware while retaining existing pricing behavior for other models.
• Recorded upstream service tiers in usage logs, with Chinese, English, and Japanese pricing hints.

Fixed

• Applied configured service tiers to Responses requests even when the client omits service_tier.
• Used the upstream-reported service tier when estimating cost, avoiding Fast overcharges when the response uses Standard pricing.

v0.0.7

[v0.0.7] - 2026-09-23

Added

• Added configurable upstream retries for transient HTTP and WebSocket failures before any response content is sent to Codex.
• Added GPT-6 Sol and GPT-6 Luna provider routes and usage pricing, including long-context rates.

Changed

• Added outbound connection settings for enabling retries, setting the retry limit, and configuring the base delay.
• Upgraded existing built-in provider catalogs with the new GPT-6 routes while preserving user customizations.

v0.0.6

[v0.0.6] - 2026-09-17

Added

• Added a repair action that detects and completes half-migrated Codex sessions, including mismatched session files, thread indexes, restore metadata, and index-only records.
• Added a localized migration confirmation dialog that explains the Codex restart and points users to the restore action if migration fails or produces an unexpected result.

Changed

• Reworked provider synchronization to stop Codex before changing session state, migrate session files and the local thread index together, reload managed configuration, and reopen Codex automatically.
• Embedded Microsoft.Data.Sqlite so session migration no longer depends on an external sqlite3 executable.
• Made session restore stop and reopen Codex so resumed conversations do not retain stale provider URLs or API keys.

Fixed

• Preserved resumable conversations when moving sessions from providers such as custom to the CodexSwitch-managed meteor-ai provider.
• Added transactional compensation and managed-file snapshots so failed provider synchronization restores session metadata, thread indexes, configuration, authentication, and model catalog files.
• Enabled migration and repair when only local thread-index records require updates.

v0.0.5

[v0.0.5] - 2026-09-15

Added

• Added an in-app help center with onboarding, feature guidance, advanced configuration, provider synchronization instructions, troubleshooting, and Chinese, English, and Japanese localization.

Changed

• Updated generated Codex model catalogs to use a 272,000-token default context window and consistently apply the optional one-million-token context setting.
• Tailored Codex reasoning-effort options to the selected model family, including max and ultra support for compatible models.

v0.0.4

[v0.0.4] - 2026-09-13

Changed

• Refined provider action controls with a consistent four-column layout and more spacious button sizing.

v0.0.3

[v0.0.3] - 2026-09-06

Fixed

• Made the default Codex desktop option launch the installed ChatGPT client through its registered protocol after the local proxy finishes initializing.

v0.0.2

[v0.0.2] - 2026-09-06

Fixed

• Replaced the remaining legacy Windows taskbar, tray, executable, and installer icon with the AIOSS application icon.

v0.0.1

[v0.0.1] - 2026-09-06

Added

• Added the refreshed CodexSwitch desktop interface, provider management, model pricing catalog, usage dashboard, Claude compatibility, and provider billing multipliers.
• Added current Claude model pricing with separate five-minute and one-hour cache-write rates.
• Added the AIOSS application and installer icon.

Changed

• Updated the default provider and model catalog, routing fallback behavior, usage accounting, and release metadata for the AIOSS distribution.