Continue CLI

You will get value on day one only if you already have this ecosystem configured, because the trait that decides it in daily use is continuity: the assistants and models you set up for the editor answer in the terminal too, with no second configuration to drift out of sync. Coming in fresh, it is a competent terminal agent among several competent terminal agents.

Pick it if the shared configuration is the thing you would miss. Pick Crush if you want a terminal agent chosen on its own merits, or Aider when you would rather steer every edit yourself.

The permission design puts everything on the invocation. Tools are gated, and the gate is lifted by a single flag on the same command line that starts the session, with per-tool allowances as the middle setting. Nothing sits behind that decision: the row records no sandboxing, and the documentation describes containers as a way to run the program rather than to confine it. One habit-forming flag is the entire boundary.

What it does right: a read-only plan mode exists as a real mode, so investigating a repository without any write path is a supported state and not a promise.

Two notes. 1. Model and assistant definitions come from a single shared schema used by every surface the project ships, which eliminates the classic divergence where a terminal client and an editor plugin disagree about what a provider is called. 2. The command-line pages defer to that shared reference instead of duplicating it, so the documentation is thin where a reader might want specifics and consistent where it matters.

No evaluation is published. The design's durability rests on that schema outliving individual providers, which is a reasonable bet and an unproven one.

The site now carries the announcement that the team has joined Cursor, with a line confirming the open code remains freely available. Read those two sentences together: the asset survives and the people who advanced it now have a different employer's priorities. Thirty-five thousand stars on the monorepo measure past distribution, not future investment.

This is the third time this cycle a well-adopted developer tool has been separated from its maintainers by an acqui-hire, and buyers keep pricing the product instead of the cap table. Position: use the binary, do not plan a migration around its roadmap, and expect community forks to fragment before one wins.

Licence cost is nothing across sixty engineers and the model spend lands on keys we issue. The part I can use is the single-shot mode, which means a pipeline step is possible and reviewable like any other script. The part that stops me is identity: the documented path is logging into a vendor account, with your own provider key as the alternative, and the row offers no single sign-on, no SCIM and no audit trail attached to either route.

Onboarding is under an hour. Approved with conditions: provider keys issued centrally, vendor accounts disabled, and a named owner for the pinned version.

Apache-2.0, no contributor gymnastics, and the code is in a monorepo I can build myself, so whatever happens to the company the binary on my disk keeps working and a fork is legally clean. Providers accept a custom apiBase, which is the entry that matters, because it means the endpoint on my own machine is configured exactly like a commercial one rather than as a special case.

It speaks MCP as a client, so the servers I already run are available without new plumbing. This is the rare tool whose ownership drama I can ignore entirely.