Upgrade from v0.1.5

This guide is for anyone moving from Melious CLI v0.1.5 to v0.2.0 who wants to know what stops working and what to use instead.

v0.2.0 narrows the CLI to one job: starting coding agents against Melious, plus one-shot prompts. We removed image generation, code review, interactive chat, and usage reporting so the launcher and run get our full attention — and because agents, the API, and the dashboard already do those jobs. If you relied on one of them, that's a real loss. Here's where each one went.

Removed commands

Removed commands and flags now fail as unknown. Conversation files you saved with v0.1.5 are left untouched — the CLI no longer reads them, but it doesn't delete them either.

One-shot melious run keeps prompt arguments, piped stdin, and file context. The launcher keeps its model picker.

Retired config fields

These fields are ignored when your existing config.json loads. They can stay in the file:

melious config get and config set on any of them report unknown config key. melious config reset would drop them, but you don't need to run it — setting other keys leaves the old fields where they are.

Behavior changes

• models --json is refused. melious models and models info <MODEL_ID> have no JSON output; drop the flag for the human table.

• Unknown commands fail. Typos and extra words exit 1 with an unknown command error and suggestions. The suggestion is a spelling match, not a replacement, so a removed command can point somewhere unrelated:

  Error: unknown command "usage" for "melious"
  Run 'melious --help' for usage.
  
  Did you mean this?
      image

• Prompts need a terminal. Bare melious, setup, config reset, update, and uninstall refuse to run without one and name the alternative — --yes for the last three, melious --no-color for a static status, auth login --stdin to sign in. See scripting.

• Codex and Pi are launchable. melious launch codex and melious launch pi join Claude Code, OpenCode, and Vibe.

• Vibe's old .env stays put. v0.1.5 wrote your key into a .env file next to Vibe's generated config. The CLI no longer writes or reads it, and doesn't delete it for you — remove it yourself if you don't want a plaintext copy of the key there.

Workspace trust

Launching an agent now requires consent for each project, so the first melious launch in every workspace asks whether to trust it. v0.1.5 had no trust records, so there's nothing to migrate — you answer once per project and the CLI remembers.

Scripts and CI without a terminal can't answer the question. Pass --trust-once to authorize a single launch without recording a grant, or run melious trust add <PROJECT_PATH> ahead of time. Workspace trust covers what a grant does and doesn't mean.

What this release doesn't include

We'd rather say this plainly than have you find out halfway through a rollout:

• No browser, device, or OAuth login. auth login takes an API key, by prompt or --stdin.
• No keychain or encrypted credential store. The key is saved as plaintext JSON, readable only by your account on Unix.
• No session accounting. The CLI doesn't total what a launched agent spent. run --show-env shows per-request data when the API returns it — that's not an account total. Totals live in the account dashboard.
• No automatic checksum verification. The installer and melious update download over HTTPS but don't verify checksums or signatures.
• No say over third-party agents. Claude Code, Codex, OpenCode, Pi, and Vibe have their own telemetry, policies, and release cycles.