OpenCode

OpenCode is the open Claude Code most people are asking for: a good terminal UI, the same agent loop shape, and you bring the model. The trait that decides it is switching. When a better model ships, you change the provider in one config line and keep your habits and keybindings, instead of waiting for a vendor to add it to a menu.

Pick it if you switch models often or already hold provider keys. Pick Claude Code if you want one vendor, one bill and nothing to configure, and Aider if you would rather the agent stay in git's lane with explicit diffs.

The risk is the absence of a sandbox. Commands run with your user's permissions in the checkout you are editing; a bad rm in a generated script is a bad rm that ran on your machine, and there is no container layer to absorb it. The failure mode is common among terminal agents, and no less serious for it.

The safe configuration is yours to build: a container, a throwaway worktree, or a permission prompt you refuse to click through. What it does right: opencode run is a non-interactive mode, so the agent can be wrapped in scripts you control and killed by a timeout you set.

Three documented properties. 1. The model layer abstracts over 75+ providers, so the loop is not coupled to one vendor's tool-calling format, which is the design most likely to survive model turnover. 2. Custom agents and subagents partition context by task, a documented way to keep a large repository's noise out of a small edit. 3. opencode serve exposes the same loop as a headless HTTP server, so a third party could drive it reproducibly.

No benchmark of any kind is published, and the docs make no claim that would require one. The observation: a star count measures attention, not resolved issues, and the documentation does not confuse the two.

The asset is attention: 203,000 GitHub stars and a license that sets switching cost at zero. The business is the Zen gateway, a per-token resale layer with several free models inside it, which is the venture-subsidized token burn I always warn about; free inference is an acquisition expense with an unpublished sunset date. The vendor renamed itself from SST to Anomaly mid-flight, which is a company still deciding what it is.

Likely acquirer: a model provider that wants an open front end with a crowd attached. Position: use the tool, pay Zen only once its unit economics are on the page, and assume the free models are temporary.

The demo is model switching in a pleasant terminal. Procurement: there is no seat, no contract and no vendor to call, which is not free at team scale. Sixty engineers with sixty provider keys is sixty spend lines, sixty retention policies set by whoever owns the key, no SSO and no audit log, and nothing central records what the agent ran.

Onboarding is an afternoon for anyone who has used a terminal agent, which is the good news. Approved with conditions: a shared gateway with spend caps, centrally issued keys that can be revoked, and a config file checked into the repo so sixty laptops run the same thing.

This is the terminal agent I actually run. MIT license, a config file with a published $schema where MCP servers sit next to provider settings, and Ollama when I am offline. The config is one JSON document, which means it lives in the repo and I can diff it, and a wrong MCP entry fails validation before it fails at runtime, which is more courtesy than most tools extend.

The Kilo CLI is a fork of it. A project that has survived a fork does not need its vendor to survive, and the fork keeps the vendor honest about what stays open. That is ownership.