Core Components

The building blocks of the ask-rb ecosystem. Each component is a standalone gem that you can use independently or compose into an agent.

Component	Purpose
LLM Providers	OpenAI, Anthropic, Google, Ollama, Bedrock, Mistral, Cloudflare
UI Kit	16 Web Components for AI chat — Lit, Shadow DOM
Tools & Execution	Tool framework, shell tools, result types
Sandbox Providers	Isolated code execution — local, Docker, Daytona, Cloudflare
The Agent Loop	Session lifecycle, think-call-execute, compaction
Attachments & File Inputs	User→agent file uploads — inline bytes or context-only manifests
Workflows & Graphs	Multi-step durable workflows with checkpointing, sub-graphs, approval, timeouts, retry
Skills	On-demand methodology for agents
Schema & Structured Output	JSON Schema DSL for tool params and structured output
Credential Resolution	Environment, file, Rails credentials, OAuth
MCP Client	Model Context Protocol client for Ruby
Web Search	Local SearXNG-backed web search — tool library and MCP server
Web Fetch	URL → clean markdown — pluggable backends: Crawl4AI, Local, Jina, Browser (launched or attached to a running Chrome)
Computer Use	Click, type, screenshot, sandboxed VMs, encrypted Computer History via Cua Driver
App Server	JSON-RPC/stdio app server — expose your agent to any app-server client
Session Protocol	Canonical wire contract for ask sessions — events, interactions, versioning
Terminal	Terminal thin client for the session protocol — the `ask` TUI
ACP Client & Server	Agent Client Protocol — JSON-RPC 2.0 over stdio: drive coding agents or host your own
RAG Pipeline	Document loaders, text splitters, vector stores, MMR — full RAG pipeline
Token Usage	Token counting, pricing, wallet engine, ledger — usage-based billing
Local Development	Stable `https://<app>.localhost` URLs — zero dependencies, worktree-aware, per-host TLS

Core Components

Table of contents