Caveman.SemanticKernel 1.4.4

🦴🔌 Caveman.SemanticKernel

Semantic Kernel plugins for Caveman — the self-contained, content-aware NLP prompt compressor for LLMs.

This is an optional add-on package. The core Caveman package has no Semantic Kernel dependency; install this package only if you want to expose Caveman's capabilities as Semantic Kernel kernel functions.

dotnet add package Caveman.SemanticKernel

Installing it automatically pulls in the Caveman core package.

Technology Partnership

<img src="https://www.digitalsolutions.it/img/partners/novaroutelogo.png" alt="NovaRouteAI" height="180" style="max-width: 100%; height: auto; min-height: 180px; max-height: 190px;">

NovaRouteAI — Build with Chinese AI models through one simple API.

NovaRouteAI helps developers and AI SaaS teams test, compare, and run models like DeepSeek, Qwen, Doubao, Kimi, and GLM without managing multiple provider accounts. Start with test credits and optimize your cost per successful task.

Click here to know NovaRouteAI

Plugins

Quick start

using caveman.core.SemanticKernel.Plugin;
using Microsoft.SemanticKernel;

var builder = Kernel.CreateBuilder();
builder.Plugins.AddFromType<CavemanConversationPlugin>();   // agent context tools
builder.Plugins.AddFromType<TokenOptimizerPlugin>();        // NLP prompt compression
builder.Plugins.AddFromType<CavemanRouterPlugin>();         // content-aware routing (v1.3.0)
builder.Plugins.AddFromType<CavemanOutputPlugin>();         // output shaping (v1.3.0)
builder.Plugins.AddFromType<CavemanRetrievalPlugin>();      // BM25+ retrieval (v1.4.3)
builder.Plugins.AddFromType<CavemanSummarizationPlugin>();  // plain-text summarization (v1.4.3)
var kernel = builder.Build();

Plugin reference

TokenOptimizerPlugin

Compresses any prompt text using Caveman's NLP engine (stop-word removal + lemmatization).

CavemanConversationPlugin

Gives a model full control over its own context window — no embeddings, no extra LLM calls.

CavemanServicesPlugin

Developer services: commit generation, code review, safety checks, stats, project exploration.

CavemanWikiPlugin

Generates AI-optimized project documentation on demand.

CavemanRouterPlugin (new in v1.3.0)

Content-aware compression pipeline: auto-detects content type and applies the best algorithm.

// Route a tool result to the best compressor
var result = await kernel.InvokeAsync("CavemanRouterPlugin", "route_content", new KernelArguments
{
    ["content"] = buildLogOutput,
    ["query"]   = "error exception",
    ["profile"] = "agent"
});

CavemanOutputPlugin (new in v1.3.0)

Reduces LLM output tokens by injecting verbosity-steering into system prompts, and protects the provider's KV-cache prefix from volatile tokens.

Verbosity levels:

// Prepare a system prompt for maximum output efficiency
var shaped = await kernel.InvokeAsync("CavemanOutputPlugin", "optimize_for_cache", new KernelArguments
{
    ["systemPrompt"]    = mySystemPrompt,
    ["verbosityLevel"]  = 2
});

CavemanRetrievalPlugin (new in v1.4.3)

BM25+ ranking over arbitrary text chunks — pure term statistics, no embeddings, no model, no network call.

// Rank log lines by relevance to an error investigation
var hits = await kernel.InvokeAsync("CavemanRetrievalPlugin", "retrieve", new KernelArguments
{
    ["documents"] = string.Join("\n---\n", logLines),
    ["query"]     = "connection timeout database",
    ["topK"]      = 5
});

CavemanSummarizationPlugin (new in v1.4.3)

Extractive summarization for plain text (use CavemanConversationPlugin.summarize_conversation instead for role-tagged transcripts).

License & attribution

Released under the Caveman License — the MIT License plus one mandatory condition:

Any use of this library must clearly and visibly disclose that it uses "Caveman" by Passaro Francesco Paolo (Digitalsolutions.it).

A note like the following, in documentation, an About screen or the repository, satisfies the requirement:

Powered by Caveman - (c) Passaro Francesco Paolo, Digitalsolutions.it

© 2026 Passaro Francesco Paolo — Digitalsolutions.it