Simple mathematical expression parser capable of generating IL code to speed up calculations.
Gives the claude CLI provider proper tool-calling: hosts the app's registered Lyntai ITools as an in-process localhost HTTP MCP server and points the CLI at it (--mcp-config), so the CLI's agent calls them natively. Opt-in via builder.AddClaudeCliMcpTools(). Uses Kestrel (ASP.NET Core) for the ephemeral, per-call, localhost-only host.
基于AScript扩展的SQL脚本语言; 快速开始: // 添加sql脚本语言 Script.Langs.Set("sql", SqlLang.Instance); string s = @" @lang sql select name,age from list where name like 'to%' and age=10 "; var script = new Script(); script.Context.SetVar("list", new [] { new Person("tommy", 15), new Person("tony", 10) }); var list = script.Eval<IEnumerable<dynamic>>(s).ToList(); Assert.AreEqual(1, list.Count); Assert.AreEqual("tony", list[0].name); Assert.AreEqual(10, list[0].age);
基于AScript扩展的JavaScript脚本语言,支持js基础语法及数据类型; 快速开始: // 注册js脚本语言 Script.Langs.Set("js", JavaScriptLang.Instance); // 执行js脚本 string s = @" @lang js function sum(a, b) { return a+b; } var n=10; sum(n,5) "; var script = new Script(); // 上下文中指定语言,或者脚本中使用@lang指定语言 //script.Context.Langs = new [] { "js" }; var result = script.Eval(s); Assert.AreEqual(15L, result);
The Lyntai backends that need nothing beyond Lyntai.Core, the BCL and Microsoft.Extensions.Http: the authenticated `claude` and `codex` CLIs (completions and self-driving agent sessions), any OpenAI-shaped HTTP endpoint (OpenAI, Azure OpenAI, OpenRouter, llama-server, LM Studio, vLLM) for chat, embeddings or reranking, Ollama's native API for chat and embeddings, and an in-process model2vec embedder — bundled because none of them ships a native payload. Media backends live in Lyntai.Generation; local GGUF inference is Lyntai.Providers.LlamaSharp; in-process transformers are Lyntai.Providers.Onnx.
Lyntai provider for in-process inference of GGUF models via LLamaSharp (llama.cpp) — no server, no port. References LLamaSharp managed only: the consuming app picks an LLamaSharp.Backend.* (Cpu/Cuda/Vulkan/Metal) matching its hardware. Wired via builder.AddLlamaSharpProvider(modelPath).
In-process transformer inference via ONNX Runtime: embeddings (a bi-encoder with the model's own pooling) or reranking (a cross-encoder over query/document pairs) — no server, no port. References the MANAGED half only — the consuming app adds one native backend (Microsoft.ML.OnnxRuntime for CPU, .DirectML for any DX12 GPU, .Gpu for CUDA), so the library never picks the user's hardware. Wired via builder.AddOnnxProvider(modelDirectory).
基于AScript扩展的python3脚本语言,支持python3基础语法及数据类型; 快速开始: // 注册Python3脚本语言 Script.Langs.Set("python3", Python3Lang.Instance); // 执行python3脚本 string s = @" @lang python3 def sum(a, b) : return a+b; n=10 sum(n,5) "; var script = new Script(); // 上下文中指定语言,或者脚本中使用@lang指定语言 //script.Context.Langs = new [] { "python3" }; var result = script.Eval(s); Assert.AreEqual(15L, result);
The dependency-free Lyntai storage backends: in-memory (UseInMemoryStorage — tests, ephemeral use, and the second backend proving the domain-interface seam) and files you can read (UseFileSystemStorage — one Markdown record per file under a root you choose). Mix either with Lyntai.Storage.Sqlite per domain via DI.
基于AScript扩展的Lua脚本语言,支持Lua基础语法及数据类型; 快速开始: // 注册lua脚本语言 Script.Langs.Set("lua", LuaLang.Instance); // 执行lua脚本 string s = @" function sum(a, b) return a + b end local n = 10 sum(n, 5) "; var script = new Script(); script.Context.Langs = new[] { "lua" }; var result = script.Eval(s); Assert.AreEqual(15L, result);
Lua io模块; 快速开始: // 注册lua脚本语言 Script.Langs.Set("lua", LuaLang.Instance); // 语言环境中注册io模块 LuaLang.Instance.AddModule("io", new LuaIOModule()); // 引入fs并调用相关方法 string s = @" require 'io' local filePath = './test.txt' local f = io.open(filePath, 'w') f:write('hello') f:close() local f = io.open(filePath, 'r') local content = f:read() f:close() "; var script = new Script(); script.Context.Langs = new[] { "lua" }; script.Eval(s);
The core library of GShell which is a REPL (Read-Eval-Print Loop) tool for Unity, supporting both Mono and IL2CPP.
ASP.NET Core integration and hypermedia primitives for Växel (Zero-eval, progressive enhancement, scoped push updates).
The sandboxed expression engine for the Tidal Report Designer: a hand-written recursive-descent parser to an AST and a tree-walking interpreter (no Roslyn, no dynamic LINQ, no eval — sandboxed by construction). Implements Core's IExpressionEvaluator with hard guards on input length, AST depth and interpreter steps.
Core .NET integration for TypeSafe AI: verification, question building, and evaluation.
C# expression interpreter/evaluator
Test-side Selenium IWebDriver for apps that embed WKWebView and host the WebViewDriver.Host eval endpoint. Maps standard WebDriver commands (find, click, sendKeys, executeScript, waits) onto the in-app JavaScript bridge, so existing Selenium page objects work on Mac Catalyst, macOS and iOS.
The oneocr command-line tool: fine-tune a recognition model from your own documents (train), measure it (eval) and work through the lines that need a human look (review).
The Sonde in-process agent: a startup hook that runs compiled C# inside a live .NET process over a local socket.
Opt-in loopback JavaScript-eval endpoint that lets WebViewDriver.Selenium automate a WKWebView (or any embedded webview) in your app. Inert unless WEBVIEWDRIVER_PORT and WEBVIEWDRIVER_TOKEN are set. Never ship this active in Release builds.
Strict-OOP, host-agnostic long-term memory library for .NET 8: scoped storage, hybrid (dense + lexical) retrieval with enforced filters, consolidation, graph retrieval, corrective recall, an eval harness, and audit-grade integrity logging — over an own C# vector-index engine (zero third-party vector dependencies; GPU acceleration via ILGPU with transparent CPU fallback included).
Pre-built bun (JavaScript runtime) native binaries. Supports Windows (x64, arm64), macOS (x64, arm64), and Linux (x64, arm64). The bun binary is automatically copied to the runtimes directory during build, and the bundled BunCli helpers locate it, fix its execute permission and run scripts or eval snippets with redirected stdin/stdout/stderr.
The tinyeval command for .NET: runs the evals in evals/*.sh against your app and says which replies pass.