Getting Started

Operator-facing guides for running Gemba against your own work items, configuring agent runtimes, and tuning parallelism.

Pages in this section

Running Gemba against your work items — the fastest path from make build to a Kanban in the browser. Covers both --beads-dir (CLI) and --dolt-url (direct-SQL) modes, Beads-only and Beads-read-only modes, expected banner output, and troubleshooting.
Agent setup — the .gemba/agents.toml schema and per-agent recipes for Claude Code, OpenAI Codex CLI, GitHub Copilot CLI, Aider (OpenAI / Anthropic / Ollama), Ollama, and plain shell.
Configuration reference — every CLI flag, environment variable, and config file Gemba reads, with the order of precedence (CLI > env > .gemba/<file> > ~/.gemba/config.toml

built-in defaults).
Install guide — includes the standard Docker image for mounted real work, the self-contained Docker quickstart image for seeded Beads exploration, and the minimal release image.
Parallelism in Gemba — declare per-agent parallelism in .gemba/agents.toml, understand intra- vs inter-session axes, the dispatcher’s try-reuse-before-spawn policy, and the SPA’s pill / global-counter surfaces.
Autonomous dispatch (pools) — quickstart — 10-minute path to a server that picks up ready beads on its own and dispatches them to sticky pooled sessions. Two flows: native (local panes) and Gas Town (rigs + polecats). Includes the SPA editor at /settings/pools.