API Reference

Generated from the source with RDoc. 74 classes and 56 modules.

Brute

• Brute — Module Brute.
• Brute::Compaction — Giving up part of a conversation so the rest still fits.
• Brute::Compaction::Middleware — Compaction strategies, as the layers of a Brute::Turn::CompactionPipeline.
• Brute::Compaction::Middleware::SlidingWindow — When rewriting the tool output was not enough, whole stretches of the conversation go.
• Brute::Compaction::Middleware::Strategy — One strategy in the ladder.
• Brute::Compaction::Middleware::ToolResults — The cheapest thing to give up first: what the tools said.
• Brute::Compaction::Summarize — Replaces stretches of the conversation with summaries of them, round after round, until it fits or there is nothing left it is allowed to give up.
• Brute::Compaction::Transcript — The shapes a compaction strategy selects over.
• Brute::Completion — Completion middlewares — the terminal step of an agent turn.
• Brute::Completion::LLMrb — Completion backed by the llm.rb gem (https://github.com/llmrb/llm.rb).
• Brute::Completion::LangChain — Completion backed by the langchainrb gem (https://github.com/patterns-ai-core/langchainrb).
• Brute::Completion::OpenRouter — Class Brute::Completion::OpenRouter.
• Brute::Completion::RubyLLM — Completion backed by the ruby_llm gem.
• Brute::Completion::RubyLLM::Tool — A Brute tool adapter wearing the interface ruby_llm’s providers read off a RubyLLM::Tool: name, description, and a params_schema (they fall back to #paramete…
• Brute::Contrib — Module Brute::Contrib.
• Brute::Contrib::LogFile — A line-oriented, append-only log file that doubles as a work queue.
• Brute::Contrib::Otel — OpenTelemetry for a turn, as hooks rather than middleware.
• Brute::Diff — Module Brute::Diff.
• Brute::Env — What the turn came back with, asked of the env itself.
• Brute::Eval — Evaluating an agent, rather than testing a middleware.
• Brute::Eval::Case — One evaluation case: what the agent is told, the world it is told it in, and what must be true of the turn afterwards.
• Brute::Eval::Suite — Runs the cases against one agent and reports what happened.
• Brute::Eval::Transcript — What one turn did.
• Brute::Eval::World — The world a case wakes up in.
• Brute::Events — @namespace
• Brute::Events::Handler — Stackable event handler base class.
• Brute::Events::PrefixedTerminalOutput — TerminalOutput variant that prefixes all output with a label.
• Brute::Events::TerminalOutput — Class Brute::Events::TerminalOutput.
• Brute::Hooks — Pub/sub registry for agent lifecycle hooks, subscribed on the builder:
• Brute::Hooks::Registry — The pub/sub registry a pipeline owns; use and run bind an emit to it.
• Brute::Hooks::Trace — The env, wrapped, for as long as one call lasts.
• Brute::MessageTransport — Class Brute::MessageTransport.
• Brute::MessageTransport::Anthropic — MessageTransport for the official anthropic gem (https://github.com/anthropics/anthropic-sdk-ruby).
• Brute::MessageTransport::LLM — MessageTransport for the llm.rb gem (https://github.com/llmrb/llm.rb).
• Brute::MessageTransport::LangChain — MessageTransport for the langchainrb gem.
• Brute::MessageTransport::OpenAI — MessageTransport for the official openai gem (https://github.com/openai/openai-ruby).
• Brute::MessageTransport::OpenRouter — MessageTransport for the open_router_enhanced gem (https://github.com/estiens/open_router_enhanced).
• Brute::MessageTransport::RubyLLM — MessageTransport for the ruby_llm gem (https://github.com/crmne/ruby_llm).
• Brute::MessageTransport::RubyLLM::V1 — ruby_llm 1.x.
• Brute::MessageTransport::RubyLLM::V2 — ruby_llm 2.0 and later.
• Brute::MessageTransport::RubyOpenAI — Class Brute::MessageTransport::RubyOpenAI.
• Brute::Messages — The in-memory conversation log is just a plain Array of Brute::Message.
• Brute::Middleware — Module Brute::Middleware.
• Brute::Middleware::Base — The parent of every middleware: it takes the next app, does its work around it, and calls it.
• Brute::Middleware::Checkpoint — Durable execution for the tool loop.
• Brute::Middleware::DefaultCompactionPipeline — Compacts the conversation once it fills too much of the model’s window.
• Brute::Middleware::DefaultCompactionPipeline::Declines — The floor of a pipeline given nothing to summarise with: what the free layers managed is what the turn gets.
• Brute::Middleware::DefaultToolPipeline — Class Brute::Middleware::DefaultToolPipeline.
• Brute::Middleware::Loop — Re-invokes the inner stack as long as a condition holds — a generic loop over a turn.
• Brute::Middleware::Loop::BackgroundJobs — Keeps the turn alive while background jobs are running.
• Brute::Middleware::Loop::ToolResult — Loops while the LLM keeps producing tool results — the standard agentic turn loop.
• Brute::Middleware::MaxIterations — Guards against runaway tool loops by capping the number of iterations.
• Brute::Middleware::Question — Class Brute::Middleware::Question.
• Brute::Middleware::SessionLog — The “session” is just a JSONL log of the conversation on disk.
• Brute::Middleware::Skills — Loads skill objects into the agent context.
• Brute::Middleware::SlashCommands — The head of every agent chain, put there by the builder itself rather than by a use anyone writes.
• Brute::Middleware::Summarize — Runs a final tool-free LLM call after the Loop::ToolResult completes, ensuring the agent produces a clean summary response.
• Brute::Middleware::SystemPrompt — Prepends a system message to env[:messages] before passing control down the middleware chain.
• Brute::Middleware::UserQueue — Class Brute::Middleware::UserQueue.
• Brute::PromptTemplate — An ERB-backed system-prompt object for Middleware::SystemPrompt — the open alternative to Brute::SystemPrompt’s built-in section stacks.
• Brute::Prompts — Module Brute::Prompts.
• Brute::Prompts::Autonomy — Module Brute::Prompts::Autonomy.
• Brute::Prompts::BuildSwitch — Module Brute::Prompts::BuildSwitch.
• Brute::Prompts::CodeReferences — Module Brute::Prompts::CodeReferences.
• Brute::Prompts::CodeStyle — Module Brute::Prompts::CodeStyle.
• Brute::Prompts::Context — Template context handed to ERB templates.
• Brute::Prompts::Conventions — Module Brute::Prompts::Conventions.
• Brute::Prompts::DoingTasks — Module Brute::Prompts::DoingTasks.
• Brute::Prompts::EditingApproach — Module Brute::Prompts::EditingApproach.
• Brute::Prompts::EditingConstraints — Module Brute::Prompts::EditingConstraints.
• Brute::Prompts::Environment — Module Brute::Prompts::Environment.
• Brute::Prompts::FrontendTasks — Module Brute::Prompts::FrontendTasks.
• Brute::Prompts::GitSafety — Module Brute::Prompts::GitSafety.
• Brute::Prompts::Identity — Module Brute::Prompts::Identity.
• Brute::Prompts::Instructions — Module Brute::Prompts::Instructions.
• Brute::Prompts::MaxSteps — Module Brute::Prompts::MaxSteps.
• Brute::Prompts::Objectivity — Module Brute::Prompts::Objectivity.
• Brute::Prompts::PlanReminder — Module Brute::Prompts::PlanReminder.
• Brute::Prompts::Proactiveness — Module Brute::Prompts::Proactiveness.
• Brute::Prompts::SecurityAndSafety — Module Brute::Prompts::SecurityAndSafety.
• Brute::Prompts::Skills — The <available_skills> system-prompt section.
• Brute::Prompts::TaskManagement — Module Brute::Prompts::TaskManagement.
• Brute::Prompts::ToneAndStyle — Module Brute::Prompts::ToneAndStyle.
• Brute::Prompts::ToolUsage — Module Brute::Prompts::ToolUsage.
• Brute::Rack — The mirror image of protocol-rack.
• Brute::Rack::Adapter — Class Brute::Rack::Adapter.
• Brute::Skill — A single skill: metadata plus the address of its SKILL.md on disk.
• Brute::SystemPrompt — Deferred system prompt builder.
• Brute::TokenCounter — How big a conversation is, in tokens.
• Brute::TokenCounter::Approximate — Tokens from text length, at a flat ratio of characters to tokens.
• Brute::TokenCounter::Rendering — The text a counter measures.
• Brute::TokenCounter::Tiktoken — Tokens from OpenAI’s byte-pair encoder, through the tiktoken_ruby gem.
• Brute::Tool — Base class for Brute’s built-in tools — a tiny, framework-agnostic tool DSL.
• Brute::Tools — Module Brute::Tools.
• Brute::Tools::Adapter — Normalizes any tool shape into one neutral interface so the rest of Brute never has to care which tools library (if any) a tool was written with.
• Brute::Tools::FS — @namespace — @namespace
• Brute::Tools::FS::FileMutationQueue — Per-file serialization queue for concurrent tool execution.
• Brute::Tools::FS::SnapshotStore — Per-path stack of file snapshots used by fs_write, fs_patch, fs_remove to enable undo.
• Brute::Tools::FSPatch — Class Brute::Tools::FSPatch.
• Brute::Tools::FSRead — Existing features (ref: opencode read tool):
• Brute::Tools::FSRemove — Class Brute::Tools::FSRemove.
• Brute::Tools::FSSearch — Existing features (ref: opencode grep tool):
• Brute::Tools::FSUndo — Class Brute::Tools::FSUndo.
• Brute::Tools::FSWrite — Class Brute::Tools::FSWrite.
• Brute::Tools::NetFetch — Class Brute::Tools::NetFetch.
• Brute::Tools::Question — Class Brute::Tools::Question.
• Brute::Tools::Shell — Existing features (ref: opencode bash tool):
• Brute::Tools::SkillLoad — The skill tool — stages 2 (activation) and 3 (execution) of the Agent Skills progressive-disclosure lifecycle (https://agentskills.io).
• Brute::Tools::SubAgent — A SubAgent is an Agent that exposes a tool-shaped facade so it can be dropped into another agent’s tools list.
• Brute::Tools::TodoList — @namespace
• Brute::Tools::TodoList::Store — In-memory todo list storage.
• Brute::Tools::TodoRead — Class Brute::Tools::TodoRead.
• Brute::Tools::TodoWrite — Class Brute::Tools::TodoWrite.
• Brute::Truncation — Universal tool output truncation.
• Brute::Turn — Module Brute::Turn.
• Brute::Turn::AgentPipeline — agent = Brute.agent # => AgentPipeline
• Brute::Turn::CompactionPipeline — A compactor built out of middleware.
• Brute::Turn::Pipeline — Class Brute::Turn::Pipeline.
• Brute::Turn::Pipeline::Chainable — Module Brute::Turn::Pipeline::Chainable.
• Brute::Turn::ToolPipeline — A ToolPipeline runs a tool call through a middleware stack.
• Brute::UsageDetection — What the provider reported, normalised.
• Brute::UsageDetection::LLMrb — llm.rb models usage as its own LLM::Usage value object.
• Brute::UsageDetection::LangChain — langchainrb has no single usage object: each provider’s response subclass answers prompt_tokens / completion_tokens / total_tokens by digging its own raw sha…
• Brute::UsageDetection::OpenRouter — OpenRouter reports usage as the provider’s raw hash on the response, in OpenAI’s wire shape.
• Brute::UsageDetection::RubyLLM — ruby_llm hangs usage off the message, not the response: a Tokens object with input/output/cache_read/cache_write/thinking, plus the provider’s own reported_c…

BruteCLI

• BruteCLI — Module BruteCLI.
• BruteCLI::Providers — Module BruteCLI::Providers.
• BruteCLI::Providers::Shell — A pseudo-LLM provider that executes user input as code via the existing Brute::Tools::Shell tool.
• BruteCLI::Providers::Shell::ModelList — Minimal object that quacks like provider.models so the REPL’s fetch_models can call provider.models.all.select(&:chat?).
• BruteCLI::Providers::ShellResponse — Synthetic completion response returned by Brute::Providers::Shell.