Channel adapter interface and implementations (Telegram, Discord, Slack) for connecting AI coding agents to messaging platforms.

A terminal-friendly interface for interacting with LLM models

DEPRECATED — maintenance mode only, superseded by the canonical ask-app-server + ask-session-protocol architecture and an internal successor. A self-hosted web coding agent for the ask-rb ecosystem. Runs Ask::Agent::Session against any workspace with a mobile-first PWA, streaming events over SSE, with approvals, plan mode, todos, and diffs. Comes with the `ach` utility CLI for headless runs and server management. Extensible to other coding agents (ACP, Codex, Claude Code) via ask-coding-providers.

Coding agent adapter interface and implementations (ACP, Codex, Claude, AskAgent) for the askoda ecosystem.

Drive desktop apps in the background via Cua Driver — screenshots, clicks, typing, sandboxed VMs, and encrypted Computer History. Wraps Cua's MCP tools as Ask::Tool subclasses and ships a bundled computer.use_computer skill.

Exposes the ask-computer tools (HistoryStatus, HistoryQuery, Screenshot, Click, Type, Key, InstallerStatus, InstallerSetup) as MCP tools. Coding agents like ZCode, Claude Code, or Cursor connect via the ask-computer-mcp stdio executable. The underlying Cua Driver remains the background daemon (cua-driver mcp); ask-computer provides the typed history and installer layer that plain cua-driver mcp does not.

Provides Ask::Provider (abstract interface), Ask::Conversation, Ask::Stream, Ask::ModelCatalog, Ask::ToolDef, Ask::Result, and structured error types. Zero runtime dependencies.

Typed questions (Choice, Score, Noul) with calibrated probabilities and confidence. Ships with a TypeSafe/Jev provider. Swap providers via the DecisionProvider registry.

Test LLM outputs with Minitest-native assertions. LLM-as-judge for faithfulness, hallucination, bias, toxicity. Deterministic assertions (contains, regex, JSON). CI-native: GitHub annotations, JUnit output, cost tracking.

Provides authenticated Octokit client, context metadata, and error guide for AI agents.

Define workflows as graphs of named steps with conditional routing, parallel execution, human-in-the-loop approval, per-item checkpointing loops, and sub-graph composition. Each step is a plain Ruby class.

Give anonymous visitors a durable identity with metered AI access: signed guest sessions, per-day turn/token budgets, single-use claiming of records into a real owner on sign-up, and stale-session retention. Framework-agnostic core with optional ActiveRecord and Rails adapters.

Error context for agents via Honeybadger for the ask-rb ecosystem.

Emits ActiveSupport::Notifications events for chat completions, embeddings, tool calls, and image generation. Works with any LLM provider. Subscribe to events for cost tracking, logging, analytics, or alerting.

Provides authenticated GraphQL client, context metadata, and error guide for AI agents working with the Linear API.

Supports OpenAI, Anthropic, Google Gemini + Vertex, Amazon Bedrock, Ollama (local), Mistral AI, and Cloudflare Workers AI + AI Gateway. One gem, all the models you need.

Railtie, install generator, and helpers wiring a Rails app into ask-local's stable .localhost URLs: host authorization, Action Cable origins, Procfile $PORT rewrite, and an agent skill for local development.

Connect to MCP servers via stdio, SSE, and Streamable HTTP transports. Run as an MCP server exposing any Ruby object as a tool. Discover tools, resources, and prompts. OAuth 2.1 authentication.

Rails engine that provides real-time monitoring for LLM usage. Cost tracking, throughput metrics, error rates, response times. Slack alerts. Real-time updates via Hotwire Turbo. Works with any ask-rb provider via ask-instrumentation events.

Provides authenticated Notion client, context metadata, and error guide for AI agents.