V0.2.0

Dashboard & API keys

• /dashboard: self-contained admin page (traffic, error rate, latency chart, account cooldowns, request table, key management)
• Per-client API keys (SHA-256 hashed, shown once on creation); existing MUSE2API_API_KEY keeps working
• Request log in SQLite (data/requests.db, 14-day retention); failed async tasks count as request errors

Images

• background="transparent" on /v1/images/generations (optional matting extra: rembg BiRefNet + pymatting)
• Reference images accepted on /v1/images/generations
• Async image tasks: async: true returns a task id, poll via GET /v1/images/generations/{task_id}

Browser driver reliability

• Parallel requests per account (own browser context, one window per tab — Chromium throttles background tabs)
• Upload approval cards auto-denied (they blocked ~10 minutes)
• Timeouts no longer cool an account down; keeps waiting up to 2.5x while muse.ai is visibly working
• Quota hint only counts in the finished agent reply (fixes healthy accounts wrongly cooled for 1 hour)
• Video generation no longer falsely reported as refused

New admin endpoints

• GET /admin/accounts/{id}/quota, /admin/keys (+/usage), /admin/requests, /admin/stats

v0.1.0

First release of the OpenAI-compatible gateway for muse.ai.

Included

• Chat, image, and video endpoints, plus account-pool admin API
• Mock driver for offline use, and a browser driver that runs muse.ai in Chromium
• Follow-up chats from the same user stay on one muse conversation and reopen its saved URL
• Browser proxy setting for networks that cannot reach muse.ai directly

Notes

• Default driver is mock. Set MUSE2API_DRIVER=browser to use a real account.
• /v1/responses, /v1/images/edits, and the direct HTTP driver are still reserved.