Repository navigation
freepro v0.1.2
Public proxy capacity under bursts
- Added four proxy sources (nine total), fetched concurrently and interleaved so a large feed cannot hide the others.
- Retain up to 6,000 candidates and rotate through them during replenishment. Background maintenance runs even when no requests arrive; blocked and busy routes no longer count as spare capacity.
- Reuse fast successful routes, using measured probe latency rather than model generation time to rank them. Each route still carries only one active request.
- Queue bursts for up to eight seconds within the existing request timeout, with a maximum of 128 waiters. A temporarily busy pool no longer immediately returns 503.
- Apply increasing cooldowns to repeated 403/429 responses and replace unavailable entries in a full pool. Catalog validation cannot override a route cooldown.
- Add ready/busy/blocked/waiting/candidate counters to
/api/statusand explanatory capacity logs. API key health remains separate from proxy capacity.
Validation: 339 local unit tests; three waves of 24 simultaneous requests through two fixture routes (72 successful requests, zero 503s); all transport fault tests; twelve successful live Muse Spark requests over more than two minutes, with 32 ready routes remaining after revalidation.
Updating
Restart freepro to run its startup update check, then choose Update. Settings, API keys and usage remain in the existing user data directory. Alternatively download the matching GUI binary below. Windows is standalone; Linux and macOS binaries need executable permission (chmod +x). macOS builds are unsigned. Verify downloads using SHA256SUMS.
Public proxies and upstream services still have finite capacity: validation does not guarantee that every generation request will be accepted, and bounded queues can fill under sustained overload.