Skip to content

Slash commands

Type / to see the whole list inline. ↑/↓ picks, tab completes. Names in brackets are aliases for the same command.

/help100×30 · 0:09
Every command with an example of it in use, and the keys, in two panels.
Putting a file back100×30 · 0:15
/undo restores the last file the agent changed, from koda's own copy — not from git.
The command palette100×30 · 0:10
Type / and the list filters as you go; tab completes.
Command What it does
/setup (/provider, /config) Set the endpoint, model and API key. Fetches the model list from whatever URL you type.
/provider <name> Switch to a saved provider and remember the choice.
/provider add Save a new named endpoint.
/model [name] Show the current model, or switch to name.
/models List the models the server reports.
/url [url] (/endpoint) Show or change the API base URL.
Command What it does
/mode [plan|execute|vibe] Show or set the mode.
/auto [ask|write|full] Cycle or set the autonomy tier.
/reason [off|low|medium|high] Cycle or set the model’s reasoning effort.
/orc <task> Orchestrate: decompose a task and delegate the parts to role agents.
/undo Put back the files the agent changed in the last turn.
/compact Summarize the conversation to free context.
/clear (/new, /reset) Drop the conversation context. Run twice to confirm.
Command What it does
/resume (/sessions) A picker of every saved conversation in this project.
/search <text> (/find) Full-text search across saved conversations.
/fork (/branch) Branch the current conversation into a copy.
/session Show which session is in play.
/copy Copy the last reply to the clipboard.
Command What it does
/websearch (/web) Turn web search on or off.
/watch Toggle watch mode: act on AI! / AI? comment triggers when idle.
/skills [reload] List skills, or reload them from disk after editing one.
/learn [accept <n>|all|reject <n>] Review rules koda learned from your usage.
/tools List the tools available right now.
Command What it does
/logs Every request, tool call, timing and failure this session.
/debug Toggle raw request/response capture to disk.
/think Show or hide the model’s reasoning.
Command What it does
/settings (/preferences, /prefs) One interactive page for everything configurable.
/theme [name] A swatch of each palette, or switch to name.
/motion Turn animation on or off.
/reveal Toggle progressive reveal of streaming text.
/mouse (/select) Toggle mouse capture. Saved to config.
/keys, /help (/?) Keys and commands.
/cwd (/pwd) Show the workspace root.
/quit (/exit, /q) Exit koda.

/motion and /reveal. /motion governs all animation — spinners, gauges, and the progressive reveal. /reveal controls only the typing-in of streaming text, and only has an effect while motion is on.

/clear and /compact. /clear throws the conversation away. /compact writes a real summary of it and keeps that, which is what you want when a session has genuinely moved on to a different task but you do not want to re-explain the project.

/mouse on and off. On, the wheel scrolls and dragging selects and copies inside koda. Off, the terminal gets click-drag selection back and you scroll with pgup/pgdn. The choice is written to the config, so it holds for every later session.

! runs the rest of the line as a shell command in the workspace — no model call, no tokens spent. $ does the same for a line of Python. Both show up in the transcript as tool blocks, but neither enters the conversation the model sees. See Command line.