Terminal AI agent gptme requires confirmation by default, ships opt-in autonomous mode

gptme, an open-source AI agent that runs from a command-line terminal, has been added to AI Dispatch's index. The project's own documentation describes it as "a great coding agent, but general-purpose enough to assist in all kinds of knowledge-work," and is distributed as "Free and open-source" software.

gptme is provider-agnostic, listing "100+ models (OpenRouter/local)" alongside direct support for Anthropic, OpenAI, Google, xAI and DeepSeek, per its own provider comparison table. It can also run entirely against local, open-weight models with no API key required.

On governance, gptme's documentation indicates tool actions require human sign-off unless disabled: its command-line help lists a flag, "-y, --no-confirm Skip all confirmation prompts," implying that confirmation prompts run by default. Separately, the project documents a "Fully non-interactive/autonomous mode (no user interaction possible, safe for scripts/CI)" intended for scripted or CI use, activated with a different flag. Which tools an agent session can use is restricted through a comma-separated allow-list — the CLI help text for that flag reads, in part, "Tools to allow as comma-separated list."

The project is also distributed as a "Desktop app for Linux (AppImage), macOS, and Windows with auto-updates," and, through the Agent Client Protocol, integrates directly into code editors: "This makes gptme available as a drop-in coding agent in Zed and JetBrains IDEs."

AI Dispatch scores agents and vendors on a five-dimension trust-gap rubric: runtime governance, audit trail, compliance, outcome settlement, and insurance/indemnity. gptme's confirmation-by-default design and scoped tool access bear on the runtime-governance dimension; its published materials make no claims about an audit trail, a compliance certification, or insurance coverage.

Full field-level sourcing for gptme is published at aidispatch.news.