v0.13.1 v0.13.2
low risk
CaviraOSS/LongMemorySep 29, 2026 → Sep 29, 20260 days between releases
What's New
Router learns model output limits to support free models with lower token caps.
New Features
- Automatic learning of model output limits from refusal responses
- Retry logic that respects dynamic output token caps
Bug Fixes
- Models with smaller output limits (e.g., Groq, Ollama) now work correctly with Claude Code
- Router retries requests within the learned output limit instead of marking models as unable
Upgrade Notes
Existing keys and profiles are retained; no database migrations are required.
The Two Releases
v0.13.1
v0.13.1
Sep 29, 2026· 23 days ago
Addresses false 'Rate limited' errors for coding agents by properly counting tool definitions to ensure each request is sent to models capable of handling them.
Full release notesv0.13.2
v0.13.2
Sep 29, 2026· 23 days ago
This release fixes compatibility for Claude Code and other agents with free models that have lower output token limits by allowing the router to learn and respect those limits from refusal responses.
Full release notes