Quickstart

Install Codag and keep working

Install the CLI, run one setup command, then use your agent as usual.

Codag connects to the coding agents you already use and lowers what they spend reading large tool results. There is no wrapper command and no new workflow. Install, run one setup command, then go back to your agent.

1. Install the CLI

The installer downloads a signed binary and verifies its checksum before it lands on your machine.

$curl -fsSL https://codag.ai/install.sh | sh

2. Run setup

$codag setup

Setup does everything in one pass:

Signs you in. Browser sign-in is the default. For CI or a headless machine, set CODAG_API_KEY to a workspace API key instead.
Proves the service works. It runs one unbilled reduction through the production service and checks the result before touching anything.
Installs a per-user local service on macOS, Linux, or Windows and waits for it to report healthy.
Attaches detected agents. Claude Code and Codex are configured automatically. Other tools are reported but never modified.
Registers a local retrieval tool so your agent can pull back exact omitted lines when it needs them.

If any step fails, setup rolls your machine back to its previous configuration. Every change it makes is reversible with codag disable or codag uninstall.

3. Use your agent normally

Open Claude Code or Codex and work as usual. Restart any agent that was already running so it reloads its configuration. From then on Codag reduces large eligible tool results before your agent reads them and passes everything else through unchanged.

4. Verify

# connection and workspace state
codag status

# measured tokens avoided and estimated dollars saved
codag savings

# rerun the production health checks
codag doctor
                        

Next: read what attach changes on the supported agents page, see every command in the CLI reference, or learn exactly what leaves your machine.