marvisx-cli 0.3.5

MarvisX OSS — agent-native company-brain CLI runtime (init, status, brief, triage)

marvis init

Interactive bootstrap wizard for MarvisX OSS. Five prompts (license, storage, BYOK, first project, recap) reusing the shared core/wizard/ state machine so the CLI and the Console /welcome route produce byte-identical settings when given the same answers.

Install

pip install marvisx-cli
marvis init

Pre-install (clone and run) works too:

python -m core.cli.marvis_init init

Quick start (interactive)

marvis init

You will be asked:

1. License — accept BSL 1.1 (y/n).
2. Storage — projects root path, database backend (sqlite or postgres), and either the SQLite file path or the Postgres DSN.
3. LLM provider (BYOK) — anthropic, openai, mac_gateway, bedrock, or skip. API key prompt is hidden input. Mac gateway asks for the base URL.
4. First project — name, slug, and type (code, work, system).
5. Recap — review choices, confirm.

On confirmation the wizard writes:

• ~/.marvis/settings.yaml — workspace + storage + llm choice (no secrets), chmod 600.
• ~/.marvis/master.key + ~/.marvis/byok.vault — Fernet-encrypted API key store (only if a provider was chosen).
• <projects_root>/<slug>/project.yaml — first project seed.

Non-interactive / CI

marvis init \
  --accept-bsl \
  --no-interactive \
  --projects-root /var/lib/marvis/projects \
  --db-backend sqlite \
  --db-path /var/lib/marvis/console.db \
  --llm-provider anthropic \
  --llm-api-key "$ANTHROPIC_API_KEY" \
  --project-name "Local Workspace" \
  --project-slug local-workspace \
  --project-type code

YAML preset (preferred for repeatable provisioning):

marvis init --no-interactive --config init.yaml

init.yaml:

welcome:
  bsl_accepted: true
storage:
  projects_root: /var/lib/marvis/projects
  db_backend: sqlite
  db_path: /var/lib/marvis/console.db
llm_provider:
  provider: anthropic
  api_key: sk-ant-...
first_project:
  name: Local Workspace
  slug: local-workspace
  type: code

Dry run

marvis init --dry-run --accept-bsl --no-interactive ...

Prints the planned filesystem writes and the rendered settings.yaml without touching the disk. Exit 0.

Exit codes

Code	Meaning
0	Success (or dry-run)
1	Generic runtime error
2	User abort at recap confirmation
3	Validation failure (field-level errors printed)

Tests

python -m pytest tests/test_cli_marvis_init.py

Ten cases cover dry-run, non-interactive flags, YAML preset, BYOK vault write, invalid slug rejection, Postgres backend, interactive prompts via CliRunner(input=...), BSL gate, and show-state JSON dump.

Relation to core/scripts/marvis-init.sh

marvis-init.sh (bash) is a deploy bootstrap: copies deploy/_template/, renders .env with secrets, runs setup-server.sh, and brings up Docker Compose. It targets server administrators provisioning a host.

marvis init (this Python CLI) is an onboarding wizard: writes settings.yaml + BYOK vault + first project seed in the user's home. It targets OSS users running MarvisX locally.

The two flows are complementary and will converge in a later phase. Until then both ship side-by-side.