v0.2.0

What's Changed

• README: link the Quest install video in Quick Start by @puntium in #19
• Bake headless LibreOffice + fonts into the sandbox images as the document converter by @cody-the-code-bot in #21
• Routine Settings: Costs section with 7/28-day, per-run and lifetime spend by @cody-the-code-bot in #22
• Animate action-card button width and routine settings dialog height by @cody-the-code-bot in #23
• Routine schedules: weekly type, restart catch-up, run ledger, quiet poll log by @cody-the-code-bot in #24
• Add Claude Sonnet 5.5 to the Vertex model registry by @cody-the-code-bot in #25
• System Reports: Models section with per-model cost, users, routine and sub-agent shares by @cody-the-code-bot in #26
• Inference Providers: self-hosted instances (OpenAI-compatible + Ollama native) by @cody-the-code-bot in #27

Full Changelog: v0.1.1...v0.2.0

v0.1.1

Highlights

Inference providers and model selection

• Admin-configured OpenRouter provider instances with their own keys and admin-picked model lists, alongside a Vertex AI card with per-model enable/disable; models carry qualified <instance>:<wire_id> ids (#8)
• New Settings > Model Selection table: composer menu slots and descriptors per model, with separate public-project slots and private/public allow flags enforced per turn (#10)
• Admin cost reports now capture and prefer OpenRouter's provider-reported per-call cost over list-price estimates (#11)

Sign-in

• Email/password sign-in as an alternative to Google OAuth, chosen per deployment via login_method; one-time set-password links for admin invites, resets and self-service (#9)

Voice input

• Composer mic button with server-side Gemini Vertex transcription, behind the new voice_input feature gate (#16)
• Waits for live audio before showing "recording", adds an input level meter, and lets Send/Enter finish a dictation in flight (#17)
• Opens the default microphone by id rather than a bare {audio: true}, so macOS no longer wakes every input device such as an iPhone Continuity mic (#18)

Fixes and performance

• Streaming state is reconciled on WebSocket subscribe, so a socket drop can no longer strand the spinner (#15)
• The composer remembers the last-used model separately for private and public projects (#14)
• Sandbox scripts start faster: crun runtime auto-detection, a single iptables-restore, and a pre-built matplotlib font cache (#12)

Full Changelog: v0.1.0...v0.1.1

v0.1.0

Initial public release of Quest.

This tag marks the state of main before upcoming refactors. Pin to v0.1.0 if you want a stable reference point.

What's included

• Initial import of the Quest codebase (#3)
• Quest logo and README branding (#1)
• Docs: readability pass, Stop vs. Deny wording, HTTPS guidance for OAuth callbacks with Tailscale Serve and Cloudflare Tunnel (#4, #5, #6)
• Link to the post on building our own harness (#7)