CLI

CLI#

gptme ships a set of command-line tools. Most of the time you only need gptme itself; the others cover setup, utilities, agents, and evaluation. Every tool accepts --help, and the reference pages list all of their options.

Chat and interfaces

  • gptme — the main CLI: chat with the assistant in your terminal.

  • gptme-server — the web UI and REST API server; see Server.

  • gptme-tui — the interactive terminal UI; see TUI.

  • gptme-mcp-server — expose gptme’s tools as an MCP server over stdio; see MCP.

  • gptme-acp — run gptme as an Agent Client Protocol agent for editors; see ACP.

Utilities

  • gptme-util — utility subcommands for models, tools, profiles, memory, and more.

  • gptme-wut — start a session with the content of the current tmux pane.

  • gptme-convert — convert files between formats (PDF, images, and more) offline.

  • gptme-attest — generate and verify output attestations.

Agents (reference)

  • gptme-agent — create and manage autonomous agents; see Agents.

  • gptme-service — scaffold a persistent headless agent as a systemd or launchd service.

Setup and diagnostics (reference)

  • gptme-onboard — first-run setup wizard.

  • gptme-auth — sign in to subscription providers and the gptme managed service.

  • gptme-init — create a new gptme project scaffold.

  • gptme-tutorial — interactive tutorial: learn gptme by doing.

  • gptme-doctor — run system diagnostics.

Sessions (reference)

  • gptme-status — generate a portable session-status document for handing off work.

  • gptme-resume — rebuild a resume prompt from a prior session.

  • gptme-checkpoint — lightweight recovery markers for Git workspaces.

  • gptme-stats — LLM usage and cost statistics across all conversations.

Evaluation and training (reference)

  • gptme-eval — run gptme’s eval suites; see Evals.

  • gptme-eval-swebench — run the SWE-bench evaluation.

  • gptme-eval-tbench — run the Terminal-Bench evaluation.

  • gptme-dataset — build fine-tuning datasets from session trajectories; see Finetuning.

  • gptme-eval-trends — track eval result trends and regressions over time (no reference page).

  • gptme-dspy — optimize system prompts with DSPy (no reference page).

Reference pages:

Common invocations#

gptme                                   # pick or start a conversation
gptme "summarize this" README.md        # start with a prompt, and a file as context
gptme -r                                # resume the most recent conversation
gptme -n "run the tests and fix failures"   # non-interactive: no confirmations
gptme --help                            # all options

See Usage for a guided tour, Commands for the slash commands available inside a session, and Configuration for configuration files and environment variables.

Contents