v0.4.1

Added

• Dedicated release workflow — pushing a v* tag now triggers a separate
  GitHub Actions pipeline that validates version alignment, builds Windows
  installers (MSI/NSIS), and attaches them to a GitHub Release.

Changed

• CI pipeline split and optimized — everyday pushes run a leaner workflow:
  Linux tests only proxy-core (~1 min), Windows builds the executable without
  bundling installers (~14 min cold cache). Full bundles are reserved for tagged
  releases.
• CI concurrency — duplicate runs on the same branch are cancelled automatically.
• Clippy runs on the Windows job (where the full crate graph is built) instead
  of a separate Linux pass.
• README documents Claude Code support (Messages API, five-slot model mapping,
  tray integration) and the split CI / release workflows.

Fixed

• tauri build --no-bundle is now passed to the Tauri CLI, not to Cargo
  (fixes a broken Windows CI build step).

v0.4.0

Changelog v0.4.0

Release date: 2026-06-28 (UTC)

Added

• Native Claude Code agent support — a third wire API (/messages, Anthropic
  Messages) joins Chat completions and Responses. Pick it via the new Messages
  segment in the endpoint switch; the UI then enables Run Claude Code and
  shows a five-slot model panel (Opus, Sonnet, Haiku, Fable, Subagent).
• Five-slot model mapping for Claude Code — each slot maps a stable
  proxy-cc/<slot> label (set in Claude Code's env at launch) to a catalog model
  id at request time. Slot choices persist per endpoint in ui_state.json. The
  subagent slot accepts a model or an Inherit toggle (mutually
  exclusive). All five slots must be configured before launch.
• Tray integration for Claude Code — on a Messages endpoint the tray shows a
  Claude Code — N/5 slots set status line (or all slots set when
  complete), one submenu per slot (with Inherit on Subagent), and Run Claude
  Code only when every slot is satisfied.
• Copilot token-limit auto-fill — when the upstream /models payload advertises
  max_prompt_tokens / max_output_tokens, those values are forwarded to Copilot
  via COPILOT_PROVIDER_MAX_* unless you override them manually in the settings
  window (per-endpoint override still persists in ui_state.json).

Changed

• The "run manually" PowerShell snippet now matches the real agent launch
  wiring (env vars and command) for Copilot, Codex, and Claude Code.
• Windows CI builds retry tauri-cli install and release builds on transient
  crates.io TLS failures and use cargo tauri build --locked for reproducible
  bundles.

Fixed

• (none)

Security

• (none)

Full commit list since v0.3.6

• feat(proxy-core): add ApiKind::Messages wire API
• feat(proxy-core): add CcSlot model-slot mapping for Claude Code
• feat(proxy-core): persist Claude Code slot config per endpoint
• feat(proxy-core): expose Claude Code slot state on AppState
• feat(proxy-core): route /messages with proxy-cc slot mapping
• test(proxy-core): cover Claude Code messages pass-through
• feat(app): add Claude Code agent launch + manual snippet
• feat(app): add set_cc_slot command and StateView slot fields
• feat(app): nest Claude Code slots in the tray menu
• feat(webview): recognize the /messages API suffix
• feat(webview): add Claude Code segment and slot panel
• docs: document Claude Code support and bump version
• chore(ci): harden Windows build against transient crates.io flakes
• style: apply rustfmt for CI fmt check
• Fix "run manually" snippet to match the real agent launch
• Auto-fill Copilot token limits from /models, with manual override
• chore(deps): bump the cargo-minor-patch group with 2 updates
• chore(ci): bump actions/checkout from 6 to 7
• chore: stop tracking CLAUDE.md (internal AI dev artifact)

v0.3.6

Release v0.3.6

v0.3.5

Release v0.3.5

v0.3.4

Release v0.3.4

v0.3.3

Release v0.3.3