0.4.6

Fixed

• Auto (router) works again. The standalone Vercel router is gone. Auto now sends Chutes' native default alias to llm.chutes.ai and, if the account has no saved routing pool, retries once with a live inline pool built from the current catalogue (:latency — fastest warm model, then failover).

Changed

• chutes.routerEndpoint default is now https://llm.chutes.ai/v1.

Install the VSIX from the assets below, or code --install-extension chutes-model-provider-vscode-0.4.6.vsix.

v0.4.5

Highlights

• Hardened model discovery, cache invalidation, streaming cancellation and tool-call validation.
• Bounded API, SSE, account and fallback processing at untrusted network boundaries.
• Reserved the virtual Auto router id and corrected quota normalization edge cases.
• Raised the minimum supported VS Code version to 1.106 for image/data chat parts.
• Documented the extension architecture, outbound request paths and endpoint validation behavior.

Security

• Malformed or unavailable tool calls now fail closed.
• Potentially expensive model-filter expressions fall back to literal matching.
• Updated vulnerable transitive packaging dependencies; npm audit reports zero vulnerabilities.

Validation

• npm run check
• 43 offline tests passed
• CI and CodeQL enabled for main
• VSIX SHA-256: DD4A7C458E6C47A49DCB384B684CBAD3DC25190DFC63B219903AA4B4D42FD1D7

The attached VSIX is the release artifact intended for VS Code Marketplace and Open VSX publication.

v0.4.4

What's changed

• Drop partially assembled tool calls when a streamed request is cancelled.
• Harden Markdown escaping for API-provided usage and quota values.
• Restore the Auto router description to English.
• Refresh release metadata and project documentation.

Validation

• CI, CodeQL, lint, type-check, formatting, 28 unit tests, production bundle, package inspection, dependency audit, secret scan, and isolated VSIX installation passed.

v0.4.3

Highlights

• Hardened model-catalogue and SSE parsing at the API boundary.
• Propagated cancellation to account requests and refreshed models immediately after setting changes.
• Rejected malformed tool arguments and escaped API-provided Markdown.
• Added TypeScript, Oxlint, Prettier, regression tests, deterministic VSIX checks, and hardened GitHub workflows.
• Added AGENTS.md, CODEOWNERS, and updated development, privacy, security, contribution, and user documentation.

Verification

• npm run check: passed
• Unit tests: 27 passed, 0 failed
• npm audit: 0 vulnerabilities
• GitHub Actions CI: passed
• CodeQL (Actions and JavaScript/TypeScript): passed

Artifact

chutes-model-provider-vscode-0.4.3.vsix

SHA-256: 03C665056DD217C4F18483E8A9EFCD0B1245628DA9DCD6BCCEE9E430FBCF836F

v0.4.2

Changed

• Improved legal documentation, trademark notices, third-party terms references, and metadata cleanup.

Tagged retroactively: this version was published to the Marketplace on 2026-07-13 but never tagged or released on GitHub.

v0.4.1

Changed

• Upgraded TypeScript from ^6.0.3 to ^7.0.0 (resolved 7.0.2). No source or configuration changes were required.

Tagged retroactively: this version was published to the Marketplace on 2026-07-10 but never tagged or released on GitHub.

v0.4.0

Auto (router) model

Adds a virtual Auto (router) model that delegates model selection and automatic fallback to Chutes' native model router. Pick it in the model list and your prompt is classified and routed to a suitable model; if that model is cold or unavailable, the router fails over automatically — no manual switching.

New settings

• chutes.autoRouterEnabled (default true) — show the Auto (router) model.
• chutes.routerEndpoint (default https://model-router-ten.vercel.app/v1) — base URL of the native router.

Install: download the .vsix below and run Extensions: Install from VSIX…, or install from the VS Code Marketplace.

v0.3.0

• New: @chutes chat participant for usage monitoring inside the chat panel.
  • @chutes /usage — spend for the current billing windows (monthly cap, 4-hour window) + daily request quota.
  • @chutes /quota — per-model quotas.
  • Reuses the API key already configured for the provider.

v0.2.1

• Fix: entering the API key could silently do nothing — saving the key fired a model-list change event mid-resolution and discarded the result. Key entry is now reliable.
• Added screenshots to the README / Marketplace listing.

v0.2.0

• Fix: selecting Chutes AI under "Manage Models" now always opens the API-key input box (previously suppressed after the first dismissal); concurrent prompts are deduped.
• Reworked README/Marketplace copy to follow VS Code chat-provider conventions; renamed the title to "Chutes AI Provider for GitHub Copilot Chat".
• Added a trademark acknowledgment.
• Removed the Open VSX badge.