ML inference backends, multi-suite scoring engine, and agent orchestrator for the Lethean AI stack. The package provides pluggable backends (Apple Metal via go-mlx, managed llama-server subprocesses, and OpenAI-compatible HTTP APIs), a concurrent scoring engine that evaluates model responses across heuristic, semantic, content, and standard benchmark suites, 23 capability probes, GGUF model management, and an SSH-based agent orchestrator that streams checkpoint evaluation results to InfluxDB and DuckDB.
Module: dappco.re/go/core/ml
Licence: EUPL-1.2
Language: Go 1.25
import "dappco.re/go/core/ml"
// HTTP backend (Ollama, LM Studio, any OpenAI-compatible endpoint)
backend := ml.NewHTTPBackend("http://localhost:11434", "qwen3:8b")
resp, err := backend.Generate(ctx, "Hello", ml.GenOpts{MaxTokens: 256})
// Scoring engine
engine := ml.NewEngine(backend, ml.Options{Suites: "heuristic,semantic", Concurrency: 4})
scores := engine.ScoreAll(responses)- Architecture — backends, scoring engine, agent orchestrator, data pipeline
- Development Guide — building, testing, contributing
- Project History — completed phases and known limitations
go test ./...
go test -race ./...
go test -bench=. ./...
go build ./...European Union Public Licence 1.2 — see LICENCE for details.