Quickstart
From first launch to an agent fixing a bug in your project, in five steps.
1. Connect a model
Run /providers, or open Preferences → Models & providers. Choose Add a provider. Pick an endpoint from the catalog, or enter a URL yourself. Then sign in or paste a key:
- ChatGPT signs in with your ChatGPT account (device sign-in).
- GitHub Copilot signs in with your GitHub account (device sign-in). It uses your Copilot plan.
- Anthropic, DeepSeek, Qwen, Ollama Cloud and custom endpoints take an API key.
- OpenAI, Google and OpenRouter take a key or an OAuth sign-in.
- Ollama Local needs nothing. It connects to
localhost:11434.
Choose Test Connection, then Save & Set as Active. Pick the model with /models (⇧⌘L).
2. Open a project
Open the project switcher in the top bar (⇧⌘O). Choose Add project, then Choose the folder Koda works in. You can also:
- Name the project.
- Initialize a git repository.
- Choose where its memory is kept.
- Choose where it runs: on this computer, or in Docker or Podman.
3. Ask for something
Type in the prompt box at the bottom of the session and press ↵:
The chunked upload test fails on the last chunk. Find out why and fix it.
The agent explores with the code graph, reads what it needs, runs commands and proposes edits. Every call appears in the transcript as it happens. Click a call to see its output.
4. Approve or let it run
On your own machine, edits and commands that change files ask first by default. An Approval required card shows the change. Press ↵ to approve it or ⌫ to reject it. You can add a note for the agent. See Approvals to change the defaults per project.
5. Review and keep the result
Changed files appear in the Changed Files panel with their line counts. Keep a change (stage it) or discard it, per file or per hunk. Run /review before you commit to get the worst problems first.
Next
- Plan larger work first with Plan mode and run it as a goal.
- Hand routine work to a cheaper model with agents or Crew.
- Learn the keyboard shortcuts: ⌘K opens the command menu.