tine run
Version 0.7 has three explicit run modes: call a bundled model adapter directly, execute a native opentine script, or wrap an external CLI harness. Each produces a portable .tine artifact.
# No-code model capture: provider or provider:model
tine run --model anthropic --prompt "Explain the current branch"
tine run --model openai:gpt-5.6 --prompt "Explain the current branch" --save first.tine
# Native script: the module must expose a Run object
tine run research.py --save research.tine
# Named CLI harnesses
tine run --harness codex --prompt "Inspect this repo" --autosave codex-draft.tine --save codex.tine
tine run --harness gemini --prompt "Summarize README.md" --save gemini.tine
tine run --harness grok --prompt "Review the patch" --save grok.tine
# Explicit generic harness
tine run --harness generic --harness-command "your-agent run" --prompt "Fix tests" --save agent.tineNo-code model capture
--model PROVIDER[:MODEL] accepts anthropic, google, ollama, and openai, plus deepseek, glm, grok, groq, hermes, kimi, ministral, mistral, openrouter, qwen, together, and zai. Without a model ID, the adapter keeps its default. No provider SDK is imported until that provider is selected.
Model mode requires --prompt and is exclusive with a script argument and --harness. Use the Python API when the run needs tools, a budget, or more than one prompt. Without --save, output goes to .tine_runs/<run-id>.tine.
Autosave
For harness runs, --autosave PATH streams atomic draft checkpoints and the harness wrapper flushes a clean final artifact. --autosave-interval N and --autosave-seconds T control throttling.
For native scripts, configure Agent(autosave_path=...) inside the script. CLI autosave flags require --harness; they are refused in script and model modes instead of silently ignored. --save works in all three modes.
Harness isolation
Harness subprocesses are isolated by default. Use --harness-login-env only for a logged-in CLI that requires narrowly scoped home, config, and path variables. Do not write pasted credentials into repository files.
Harnesses and bounds
Named presets include Claude Code, Codex, Cursor, Gemini CLI, Grok Build, Hermes, Kimi Code, OpenClaw, OpenCode, and Pi, plus the generic surface. Pi and generic require --harness-command. Presets encode documented non-interactive invocations; they do not imply every upstream CLI release has a live compatibility gate.
External processes default to a one-hour timeout, four-million-character retained output ceiling, 10,000 parsed events, and a one-million-byte line ceiling. Adjust them with the --harness-timeout, --harness-max-output, --harness-max-events, and --harness-max-line-bytes flags.