v0.11.0

FreeLLMAPI v0.11.0 reworks the Playground around a liquid composer with voice dictation and an artifacts side panel, folds the Providers page into Keys, and fixes the failover chain burning its budget on stale models and empty keys.

## Playground

- **Composer.** The input bar and the send button are one body of liquid: the button emerges from under the bar's right edge once there is something to send. Muted fill, no border, and the chat sits in a centred column.
- **Dictation.** A mic in the bar records and transcribes through the gateway's own `/v1/audio/transcriptions` route, so it uses whichever transcription models have keys. Live level bars show while you speak. Pick Auto or a specific transcription model under Settings. With no transcription model configured, the mic explains what to add.
- **Artifacts.** HTML and SVG code in a reply renders as a card that opens a side panel: a sandboxed live preview next to the highlighted source, with reload, copy and download. The panel slides in and out, resizes from its left edge, and remembers its width. (#1245)
- **Replies.** Assistant messages are full-width text with no bubble. A small maker icon (Google, Meta, Qwen, DeepSeek, Mistral and others) inferred from the model id replaces the provider name, with the provider in the tooltip. Reasoning traces read at body size. Code blocks are syntax highlighted.
- **Sidebar.** New chat is a proper button, conversation rows show titles with a hover menu for rename and delete, the chat header is gone, and the panes use lighter dividers and icons.

## Dashboard

- **Keys absorbs Providers.** The Providers page is removed; its two unique features moved into Keys. Every provider's menu now has *Test models*, which fires one real request per model through the normal routing path and shows pass or fail with latency, and *Add Custom Model* for hand-typed model ids. Custom endpoint keys get both on the key row. (#1244)
- **Monthly token budget** on the Models page is one flat list of models, smartest first, with no provider grouping. (#1243)

## Routing

- A provider answering "The model 'x' does not exist" is now treated as a stale model and skipped for the rest of the request instead of burning one failover hop per dead row. (#1242, closes #1239)
- A 402 out-of-credits response benches the key on every model of that provider for the credit window, not just the one model that failed.
- The 400 "All routed providers rejected the request as invalid" body only appears when every attempt was a genuine bad request. Mixed trails return a 502 with a per-class breakdown such as `provider_bad_request ×5, out_of_credits ×3`.

## Updating

Download the installer for your platform below, or use `ghcr.io/tashfeenahmed/freellmapi:v0.11.0` for Docker. The `latest` Docker tag also points to this release.

Existing keys and profiles are retained. No new database migrations in this release. Back up your data before updating a self-hosted instance.

[Full changelog](https://github.com/tashfeenahmed/freellmapi/compare/v0.10.1...v0.11.0)

---
⭐ Like the free router? [Go Premium](https://freellmapi.co/?utm_source=github&utm_medium=release&utm_campaign=premium&utm_content=v0.11.0#pricing), the live signed catalog, $19/yr, cancel anytime.