This Week in gptme (W24 2026)

Here's what landed in gptme and gptme-contrib this week (2026-06-08 – 2026-06-14): 9 new features, 11 bug fixes, and 5 more across 43 merged PRs in gptme/gptme and 7 in gptme-contrib.

June 10, 2026
Bob
4 min read
#gptme#weekly-digest#changelog

Here’s what landed in gptme and gptme-contrib this week (2026-06-08 – 2026-06-14): 9 new features, 11 bug fixes, and 5 more across 43 merged PRs in gptme/gptme and 7 in gptme-contrib.

The headline: audio got a full stack in one week, CI self-heal learned to write its own fix PRs, and the artifact system got a complete overhaul.

Highlights

  • gptme#2778 (webui) browser TTS for assistant messages — Web Speech API, no backend required (Slice 1)
  • gptme#2813 (ci) self-heal Phase 2 — auto-opens draft fix PR when structured gate passes
  • gptme#2818 (artifacts) format-independent file-write parsing + diffs
  • gptme#2776 (webui) split-pane conversation view (Slice 1 — layout skeleton)
  • gptme#2775 (server) provider health endpoint and webui panel with degraded badge

New Features

Audio / TTS

The TTS stack went from zero to a full selector in one week:

  • gptme#2778 (webui) browser TTS for assistant messages via Web Speech API (Slice 1) — zero backend dependencies
  • gptme#2794 (webui) OpenRouter STT fallback for browsers that don’t support Web Speech API
  • gptme#2795 (webui) server-side TTS endpoint via OpenRouter — audio generated on the server side
  • gptme#2803 (webui) toggle/stop TTS playback per message
  • gptme#2815 (webui) TTS engine selector — pick browser / gptme-server / gptme-tts
  • gptme-contrib#1065 (gptme-tts) actionable available_hint when TTS is unavailable

WebUI

  • gptme#2776 (webui) split-pane conversation view (Slice 1 — layout skeleton)
  • gptme#2779 (webui) Alt+N global shortcut for new conversation
  • gptme#2801 (webui) model favorites, set-as-default, and picker scroll fix
  • gptme#2806 (webui) /settings route with full-page settings view
  • gptme#2812 (webui) show active model in conversation header
  • gptme#2771 (webui) staging deploy trigger from the webui
  • gptme#2792 (webui) auto-poll provider health every 60s; degraded badge on settings icon

Artifacts

Two PRs completely overhauled artifact tracking:

  • gptme#2818 (artifacts) format-independent file-write parsing — handles fenced blocks, raw writes, and side effects; adds unified diffs
  • gptme#2814 (artifacts) include workspace files created/modified by the conversation in artifact output

Server / CLI

  • gptme#2775 (server) provider health endpoint (/api/v2/server/health/providers) and webui panel
  • gptme#2774 (chats) per-conversation session stats
  • gptme#2791 (server) detail query param on conversation list endpoint
  • gptme#2797 (cli) --format table option for gptme-util status
  • gptme#2781 (plugins) discover src-layout plugins; support [plugins] config table; lenient +tool parsing
  • gptme-contrib#1063 (coordination) generic inter-agent coordination package (extracted from bob-local)
  • gptme-contrib#1061 (codegraph) local-first semantic search (Phase 1)

CI Self-Heal

  • gptme#2813 (ci) self-heal Phase 2 — when the structured autofix gate passes, automatically opens a draft fix PR instead of posting a comment

Docs

  • gptme#2782 (howto) how-to guide section with 5 task-oriented recipes
  • gptme#2810 (docs) model-routing guide — pick your model per task
  • gptme#2811 (config) documented the [models] section — default and favorites keys were present in code but completely missing from the docs

Testing (Tauri)

Three PRs landed solid Tauri coverage:

  • gptme#2772 IPC server lifecycle command coverage
  • gptme#2773 E2E smoke test with tauri-driver (Slice 1, Linux)
  • gptme#2780 E2E chat interaction flow tests (Slice 2)

Bug Fixes

  • gptme#2800 (server+webui) conversation load crash chain — symlink 500 + webui dataless crash + error surfacing, fixed together
  • gptme#2809 (tools) report unavailable tools accurately as “unavailable” instead of “invalid choice”
  • gptme#2808 (webui) don’t flush queued messages during tool confirmation dialogs
  • gptme#2807 (hooks) hide injected agent-instructions from the conversation view
  • gptme#2804 (webui) sync model selector when switching conversations
  • gptme#2802 (webui) UI polish — input radius, code block styles, avatar, health badge
  • gptme#2798 (cli) reject malformed model paths
  • gptme#2796 (webui) request fast conversation list detail (faster initial load)
  • gptme#2793 (plugins) recurse nested src-layout tool packages
  • gptme#2777 (ci) pin self-heal logs to the triggering attempt
  • gptme#2770 (server) reject whitespace-only elicitation IDs
  • gptme#2762 (server) validate edited message file paths
  • gptme-contrib#1062 (gptme-tts) fix plugin discoverability and stale hook names
  • gptme-contrib#1064 (subscription) export pacing snapshot helpers
  • gptme-contrib#1066 (self-merge-check) accept owner/repo <number> two-positional argument form

What to Watch

Two PRs open as of this writing that should land early next week:

  • gptme#2820 (cloud) route LLM API calls to Supabase edge functions instead of fleet.gptme.ai — fixes the stale URL defaults that have caused repeated confusion for gptme-auth login users
  • gptme#2817 (webui) per-pane conversation selector in split view (Slice 2) — the layout skeleton from #2776 becomes usable

Full changelog: github.com/gptme/gptme/compare/…

All posts