tura

tura [globaloptions] [prompt | command [args_...]]

tura is an open-source agent runtime harness designed to complete long-horizon coding tasks with fewer model turns and tokens than typical tool-calling loops. Instead of exposing many fine-grained tools per step, it emphasizes a macro command_run tool so related shell, patch, build, and test actions can run in one structured turn, plus runtime context management tied to task state.The primary entry point is the tura command: with no arguments it launches a terminal UI; tura run and tura exec provide non-interactive paths for scripts and CI-style prompts. A local gateway process coordinates sessions, agents, and providers. Install via npm (tura-ai) or from source with the project's install scripts; on first launch, configure an LLM provider and model before sending prompts.Published benchmarks compare Tura agent configurations against other coding agents on long-horizon tasks; results and methodology are documented upstream. License is AGPL-3.0-or-later.

--gateway-url URL

Use an explicit gateway instead of auto-starting or discovering one.

Workspace directory sent to the gateway. Defaults to the current directory.

Open the TUI on a specific session. Also read from TURA_TUI_INITIAL_SESSION_ID.

Request JSON output where the selected command supports it.

Print gateway request diagnostics to stderr.

Force plain/safe terminal rendering or rich rendering.

Set CLI display language.

run [options] "prompt"

Gateway-backed non-interactive prompt: create or reuse a session, stream until the turn completes, print the result. Options include -m/--model, -a/--agent, --session, --output text|json|ndjson, --timeout, --stream/--no-stream, --bash/--zsh/--shel.

Direct Rust CLI one-shot runner (tura_exec). Reads prompt from stdin if omitted. Options include -C/--cwd, -m/--model, -a/--agent, --json, --quiet, --sandbox, --goal, --bash/--zsh/--shll.

Aliases for run with the command_run shell surface forced.

Inspect and modify gateway sessions. list supports --all and --json.

Show a session, or append a follow-up prompt.

Read or patch workspace session config and model tiers.

List providers and manage local authentication (OAuth or API key). Credentials are not bundled; configure a provider before prompt commands.

Work with the gateway agent registry (default agent id often balanced).

Work with the gateway persona registry.

Inspect or create gateway workspaces.

Operate on files in the selected workspace via the gateway.

List or execute registered gateway commands (not arbitrary shell).

Diagnostics for gateway and session state.

Print shell completion snippets.

Send one raw HTTP request through the CLI gateway client.

npm package-entry helpers to register or check the release binary on PATH.

Requires a configured LLM provider; installation alone does not supply credentials. Behavior and available models depend on which providers you authenticate. Some subcommands need a running or auto-started gateway. Companion binaries (tura_gateway, tura_gui, router/session DB tools) are for service and desktop use, not everyday prompt entry. Spelling of shell surfaces differs slightly between TypeScript (shel) and Rust (shll) frontends.

Developed by Tura-AI as an open-source agent harness with TUI, CLI, and optional GUI, emphasizing macro command execution and context control. Distributed on npm as tura-ai and on GitHub as Tura-AI/tura.

codex(1), aider(1), opencode(1), npm(1)

Source code · Homepage · Documentation