FreeLLMAPI v0.13.0 fixes Claude Code on Gemini, makes the Windows desktop app visible in the tray, fills model lists for built-in providers the catalog does not carry, and handles Gemini's daily quota properly.
New
- Fetch models for more built-in providers. SiliconFlow, LongCat and GitHub keys could pass the health check while serving zero models. Their keys now get Fetch models in the dashboard, which reads the provider's own model list and lets you pick which ones to use. Picked models are marked "discovered", and the signed catalog always takes priority over them. Set
BUILTIN_MODEL_DISCOVERY=autoto fetch them automatically. (#1357, [#1359], closes [#1348]) - Optional keys for AI Horde, Kilo and OVH. These providers work without a key, and now you can also save a real key for them to get your own account's limits. Custom providers without a key no longer send a placeholder
Authorizationheader. (#1356, [#1332], closes [#1331]) - Monthly usage and budget caps in the edit-key dialog. See this month's requests and tokens for each key and set a monthly request or token cap. (#1351)
Fixes
- Claude Code on Gemini. Tool schemas with nested or tuple arrays were rejected by Gemini with
items: missing field, so every Gemini model failed on the first message. Every array now gets a schema Gemini accepts. (#1338, closes [#1334] and [#1349]) - Gemini daily quota. When Gemini reports its daily quota is used up, that route now waits until the midnight Pacific reset instead of retrying every 90 seconds, and requests move on to your other providers. (#1343, thanks @BNTiyan)
- Windows tray icon. The tray icon failed to appear on Windows, so the app looked like it had crashed. It now uses a proper Windows icon, left-click opens the dashboard, and the dashboard opens once on first launch. (#1355, closes [#1353])
- Requesty's free tier is 200 requests a day shared across all its models, and is now capped that way. (#1328, thanks @Thibaultjaigu)
- Log feedback loop. A closed stdout (for example a detached terminal) could make the server write errors about writing errors, endlessly. (#1345, thanks @chongjiazhen)
- "No models" notice for Speechify and other audio or embedding providers no longer shows on every key, and Premium installs are no longer told to buy Premium. (#1329, closes [#1327])
- Completions always include a usage block, estimated when the provider leaves it out. (#1336)
- Login lockout and reset-code cooldown responses now include
Retry-After. (#1341)
Updating
Download the installer for your platform below, or use ghcr.io/tashfeenahmed/freellmapi:v0.13.0 for Docker. The latest Docker tag also points to this release. Desktop users on v0.12.0 will see this update in the Updates panel.
Existing keys and profiles are retained. This release has no database migrations.
⭐ Like the free router? Go Premium, the live signed catalog, $19/yr, cancel anytime.
