static
Open source projects
Repos and live builds. The full why, problem, and result for each flagship is on the home page.
-
py-bug-trace · Laguna eval
12th of 29 overall, and 2nd on the hardest level: weak on easy tasks, strong at fixing code.
-
UK Charity Doc Extract
122 scored runs. The best models reach about 0.95 F1, and the leaderboard shows the failures too.
-
AgentVetter
Scans agent installs and blocks them before they run. Tiered routing uses SIE before a heavier model step.
-
SIE
- Why
- AgentVetter and the RAG parameter toolkit need a shared place to run models.
- Problem
- Embeddings, routing, and triage need an inference server those projects can call.
- Built
- Superlinked Inference Engine (SIE) — an open-source inference server and production cluster.
- Result
- AgentVetter uses it for triage before a heavier model step. rag-params-finder uses it for embeddings.
SIE-enabled dependent projects- AgentVetter — sandboxed security scanning for AI skills and MCP servers; SIE handles tiered triage before Model Studio escalation.
- rag-params-finder — RAG parameter sweep toolkit (embeddings, chunking, retrieval); supports SIE open-source embeddings via remote gateway or self-hosted Docker.