From repository to control center.

The installer finds the newest macOS or Linux release, verifies its checksum, stages and validates the binary, then places it in ~/.local/bin.

$ curl -fsSL https://raw.githubusercontent.com/kaltstart-co/agentklar/main/install.sh | bash -s --

Choose another install directory when needed:

$ curl -fsSL https://raw.githubusercontent.com/kaltstart-co/agentklar/main/install.sh \
  | AGENTKLAR_INSTALL_DIR="$HOME/bin" bash -s --

Run init once inside every repository you want in the catalog. It creates or reuses that project's workspace and proposes its quality configuration.

$ cd /path/to/your-project
$ agentklar init
$ agentklar ui --open

The UI is multi-project. Register projects from their repositories, then switch among all of them from the same control center.

Install Agentklar's skill and commands for the supported clients. Then print the exact MCP configuration for the agent you want to connect.

$ agentklar install --agents opencode,claude,codex
$ agentklar mcp install --client codex

# also supported: opencode, or omit --client for all snippets

A task needs acceptance criteria and a verification method before it can become Ready. This is the minimum contract an agent receives.

$ agentklar task new AK-1 "Fix the parser" \
  --criteria "handles empty input;tests pass" \
  --verify "go test ./..."
$ agentklar task ready AK-1

Let the connected agent claim and submit the task. After submission, run agentklar gate <id> to execute the project's declared recipes and store the evidence. A person handles the final decision from the Approvals view.

The install command is also the update command. It validates the new binary before replacing the installed one.

$ curl -fsSL https://raw.githubusercontent.com/kaltstart-co/agentklar/main/install.sh | bash -s --

Your project data stays in place. Updating replaces the binary and its embedded UI. The catalog and per-project SQLite workspaces remain under the Agentklar data directory.