Python SDK for LoRA fine-tuning with supervised fine-tuning and reinforcement learning
Stage
Startup
Not given
Founded
2025
Not given
Based in
CA
Not given
Pricing model
Freemium
Not given
Pricing
Free: $0/month with 100 API calls/month, 1 seat, basic analytics, 30-day retention. Pro: $49/month with unlimited API calls, 5 seats, full analytics, hard spending caps, 90-day retention.
Not given
Key features
Real-time cost tracking with per-model and per-tag breakdowns
Budget alerts via email, Slack, or Teams at configurable thresholds
Hard spending caps that block requests with 429 response
Rate limiting with sliding window per API key
Support for 9 LLM providers and 60+ models
Analytics dashboard with daily spend charts and latency metrics
Works with any programming language via HTTP proxy
AES-256 encryption for provider API keys with BYOK model
90-day data retention on Pro plan
Tag-based cost attribution by team, feature, or customer
forward_backward() operation for computing loss and accumulating gradients
optim_step() for updating the adapter
sample() for generating from current or saved policy
save_state() to save and resume training state
Tinker API compatibility
Separate trainer and sampler resources
Synchronous and asynchronous RL patterns
Frozen base model weights with LoRA adapter updates
Support for open models from Qwen, Google, DeepSeek, Moonshot AI, NVIDIA, and Z.ai
What makes it different
Minimal 31ms proxy overhead, only 3.6% latency impact on typical calls
No SDKs required, works with any HTTP client out of the box
Never stores prompts or completions, only billing metadata
Hard spending caps enforced by Redis to prevent overspend
Managed training and sampling infrastructure, no setup required
Tinker-compatible API for easy migration
Ready-made SFT and RL recipes through Loops cookbook
Tokonomics
Loops