TallpineTallpine Docs

Agent adapters

The agent harnesses built into Tallpine, with terminal and structured Chat capabilities.

View on GitHub

Tallpine supports four agent harnesses: Claude Code, Codex, Kiro and DeepSeek. Each needs a local installation and sign-in. For an installed agent, check that Tallpine can find it and start a session:

tallpine agent ls --refresh
tallpine spawn --project my-app --name task --agent codex

Tallpine uses the agent CLI and authentication you already installed. It does not bundle external provider credentials or install external adapters from npm.

Shipped agent harnesses

The current daemon registers the following capabilities. Availability still depends on platform, installed version, credentials, and permissions; a registered adapter is not proof that it is ready on your machine.

HarnessTerminal UIChatReviewer
claude-codeYesYesYes
codexYesYesYes
kiroYes—Yes
deepseek-harnessYes (one-shot task)Yes—

The table shows shipped capabilities, not whether an agent is ready on your machine. Run the catalog for current readiness or when scripting:

tallpine agent ls --refresh
tallpine agent ls --refresh --json

The readiness probe reports whether a harness is supported, installed, and authenticated. Spawn still performs authoritative validation.

Chat versus Terminal UI

Claude Code, Codex and DeepSeek have structured Chat drivers:

HarnessChat transport
codexCodex native app-server
claude-codeclaude-agent-acp using the user's Claude executable
deepseek-harnessdsh Agent Client Protocol (ACP) server

Kiro is Terminal UI-only. Choose the initial interface with tallpine spawn --mode chat|tui. Compatible Claude Code and Codex sessions can hand off between Chat and Terminal UI while retaining the Tallpine session and worktree.

Chat registration does not imply interface handoff support. Use the switch action and its disabled reason in the session UI.

Choosing and configuring a harness

tallpine spawn --project my-app --name auth-fix --agent claude-code --mode chat

To make Codex the default worker, use tallpine project set-config my-app --worker-agent codex. Focused flags merge into the existing config; only --config-json replaces it. See Update from the CLI.

--agent and --harness are aliases. The CLI's project-config mirror also omits effort: tallpine project get --json cannot show it, and a CLI configuration replacement drops existing effort values. Use Project Settings in the desktop app or the daemon API to preserve or change effort. Use tallpine project set-config --help for supported fields.

Harnesses that were removed

Earlier versions supported more harnesses. A project, automation or reviewer setting that still names a removed harness falls back to Claude Code (a warning is logged), and tallpine spawn prints the same notice. A session that was created under a removed harness stays visible but is read-only: it cannot be restored, resumed, messaged or switched, and the session view explains why. Terminate it, or start a new session with Claude Code, Codex or Kiro.

On this page