Documentation

Basics

Getting StartedInstallation, first run, what happens under the hood
UsageOne-shot mode, REPL mode, CLI flags, piping, exit codes
ToolsFile ops, search, editing, web fetching, thinking, command execution
Context ManagementHow Swival fits large tasks into small context windows
Not Just for Frontier ModelsWhy Swival is built to work well with small and open models too

Safety & SandboxingPath resolution, symlink protection, command whitelisting
Secret EncryptionFormat-preserving encryption of credentials before they reach the LLM provider
Outbound LLM FilterUser-defined scripts that redact or block content before it reaches the LLM provider
Command MiddlewareIntercept, rewrite, or block shell commands before execution — RTK integration included
SkillsCreating and using SKILL.md-based agent skills
Web BrowsingChrome DevTools MCP, agent-browser, and Lightpanda
MCPConnecting external tool servers via Model Context Protocol
A2AConnecting to remote agents via the Agent-to-Agent protocol
Lifecycle HooksRun commands at startup and exit to sync .swival/ state with remote storage
Custom CommandsRun external scripts from the REPL and inject their output into the conversation
CustomizationConfig files, project instructions, system prompt overrides, tuning parameters
ProvidersLM Studio, HuggingFace, OpenRouter, Google Gemini, ChatGPT Plus/Pro, AWS Bedrock, and generic server configuration
ReportsJSON reports for benchmarking and evaluation
ReviewsExternal reviewer scripts for automated QA gates
AgentFSCopy-on-write filesystem sandboxing