#113113th of 142 in Agent frameworks and SDKs
semif-openjev
Reads typed option probabilities from a local LLM instead of generating text
- Stars
- 4.9k
- License
- MIT
- Last commit
- Sep 2026
- Language
- Python
75 on 16 Sept 20264,949 on 11 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
SemIf scores runtime-defined decisions (routing, retry, evidence checks) by reading the logits of declared options in one forward pass, with no sampled answer or JSON parsing. It runs on CUDA, Apple Silicon (MLX or PyTorch/MPS) and CPU through llama.cpp GGUF, and can reuse a shared state across many criteria. It reproduces the interface pattern of TypeSafe's closed Jev service, not its model.
Who it is for: Engineers wanting fast local semantic decisions inside agent code
Strengths
- Direct logits took 1.023 s versus 5.332 s for a JSON array on 21 criteria
- Shared-state mode reached 20.03 decisions/s on a 37x21 workload versus 2.33 fresh
- Backends for CUDA, MLX, PyTorch/MPS and CPU-only llama.cpp GGUF
- Fixtures, row-level outputs, prompt hashes and model revisions are committed, plus per-workload calibration
Weaknesses
- Prefix-reuse paths are experimental; BF16 changed 5-6 of 777 argmaxes
- 4B baseline reaches 0.845 agreement on the TypeSafe subset versus 0.883 for Jev
- Probabilities are conditional on the supplied options and need calibration per workload
- Research baseline with no release and no server or API
What it needs
- GPU optional
- Needs PyTorch, Hugging Face model weights, llama.cpp (CPU backend), MLX (Apple Silicon backend)
- Models: Qwen3.5-4B, MiniCPM5-2B, Qwen3-0.6B, Qwen3.8-27B EXL3, GGUF (llama.cpp)
Also in Agent frameworks and SDKs
See all 142| Rank | Project | Score |
|---|---|---|
| 1 | ppt-masterAgent skill that turns documents into natively editable PowerPoint decks | 96 out of 100 |
| 2 | openai-agents-pythonPython SDK for multi-agent workflows with handoffs, guardrails and tracing | 92 out of 100 |
| 3 | mastraTypeScript framework for building AI agents, workflows and MCP servers | 92 out of 100 |
| #4 | browser-usePython library that lets an LLM agent control a web browser | 90 out of 100 |
| #5 | copilotkitReact and Angular SDK for adding agent chat and generative UI to apps | 90 out of 100 |