v1.2.1

[1.2.1] - 2026-10-20

Fixed

• API streaming: Stream incremental response deltas and emit protocol-formatted errors when the upstream stream fails.
• API request validation: Reject request bodies over 10 MiB and return validation errors without echoing submitted values.
• API client cache: Cap cached clients per token and expire inactive clients.
• Conversation concurrency: Serialize operations on cached threads and move blocking requests off the FastAPI event loop.
• HTTP sessions: Reject absolute URLs in authenticated requests; serialize access to the shared curl-cffi session; validate retry, timeout, and rate-limit settings.
• Account checks: Apply the request limiter to session and settings lookups.
• Thread continuation: Reject attempts to change models on an existing thread.
• Search settings: Map source_focus="all" to web and scholar; map writing mode to search_focus="internet" with search disabled.
• Container CI: Classify container build inputs, build API and MCP images, and smoke-test both CLI entry points.
• PyPI recovery: Fail if the requested version already exists. Verify artifact digests before resuming an interrupted publish.

Changed

• Model catalog: Added current WebUI models and updated availability metadata.
• Optional dependencies: Renamed [cli] to [token]. [all] installs [token], [api], and [mcp].
• API: Added /v1/responses and retained /v1/chat/completions. Server-configured upstream tokens require a separate gateway key on each request.
• CLI: Removed terminal chat and saved-token storage. Token generation, API, and MCP remain separate optional commands.
• CLI options: Added -h and --help to all commands, plus -e for the token command's --email option.
• Python support: Set .python-version to Python 3.15, added it to the supported range, and included it in the CI test matrix.
• Request payloads: Added WebUI fields for new conversations and follow-up prompts.
• Documentation hosting: Clarified that Read the Docs publishes separately from package and container releases.