Glossary

Agent — A configured conversational entity with a specific system prompt, toolset, and permission profile. Ptah ships with a catalog and you can add your own under .claude/agents/.

Autopilot — A set of toggles that let an agent perform actions without pausing for approval. Off by default. See Settings → Autopilot.

Effort level — A hint to the provider about how much reasoning to spend on a request. fast, balanced, or deep. Providers that don’t support reasoning budgets ignore it.

MCP — Model Context Protocol. The standard Ptah uses to expose tools to agents. Ptah ships a built-in MCP server and can connect to external ones.

Orchestration — A workflow mode where a lead agent breaks work into tasks and delegates to specialist sub-agents. Pro-tier.

Plugin — A bundle of agents, skills, and tools that extend Ptah for a specific domain. Plugins live under ~/.ptah/plugins/ and are enabled per workspace via .ptah/plugins.json.

Provider — A source of language-model completions (Claude, Copilot, Codex, Gemini, Ollama, OpenRouter). Configured in Settings → Providers.

Session — A single chat conversation, saved to disk and resumable. Sessions belong to a workspace.

Skill — A named, reusable procedure with its own instructions and optional allowed tools. Invoked from chat. Lives in .claude/skills/ (workspace) or ~/.claude/skills/ (global).

Template — A scaffold for a new project, plugin, or skill. Templates live under ~/.ptah/templates/.

Worktree — A secondary working copy of a Git repository. Ptah can create and manage worktrees so an agent can experiment without disturbing your main branch.

Workspace — The root folder of a project, as Ptah sees it. Settings under .claude/ and .ptah/ in that folder apply.