kiri-cli

Kiri is the command-line interface for Kirimana — a source-available, AI-driven, platform-agnostic data-platform automation tool. FSL-1.1-ALv2 (converts to Apache-2.0 after two years).

This package ships a single kiri binary that drives the full data-contract lifecycle: scaffold a new data warehouse project, declare sources + contracts, generate dbt models, apply against DuckDB / Databricks / Fabric / Trino / Iceberg, govern with RBAC + audit + AI policy, ship to production with multi-environment release manifests.

The internal package mappers (packages/cli/kiri_cli/, kiri_core, etc.) and ODCS customProperties (kiri.classification, etc.) are named kiri.* — those are stable de-facto contracts. Kiri is the user-facing surface; Kirimana is the implementation namespace. See the deployment docs for the full taxonomy.

Install

pip install kiri-cli                   # base + localduckdb adapter
pip install kiri-cli[databricks]       # + Databricks (most common pilot)
pip install kiri-cli[trino]            # + Trino (OSS lake stack)
pip install kiri-cli[iceberg]          # + Apache Iceberg
pip install kiri-cli[all]              # all of the above

Recommended: install via uv so the binary lands in ~/.local/bin/:

uv tool install kiri-cli

Or via pipx:

pipx install kiri-cli

Both isolate the install in its own venv so it doesn't collide with your project's Python environment.

Quickstart

Build a real, populated warehouse table — no cloud account, no credentials, nothing to author:

kiri --version
kiri init my_warehouse --with-sample-data --silver-technique flat
cd my_warehouse
kiri plan                           # preview what will be built
kiri apply                          # ingests the bundled sample into bronze

--with-sample-data drops a small CSV plus its data contract into the fresh project, so apply has something real to ingest and you end up with a queryable DuckDB table on the first run. Drop the flag when you point it at your own sources.

Two things worth knowing:

• --silver-technique is required and has no default. It is an expensive choice to reverse, so the CLI makes you state it rather than guessing (flat is the right answer while you are exploring).
• If your machine has no git identity yet, init still creates and activates the project — it stages every file, skips the initial commit, and prints the git config commands to finish it.

Next: kiri doctor for a health punch-list, kiri onboard to brief your LLM assistant on the whole product, and kiri --help for every command group (kiri <group> --help drills in).

Top-level command groups

Source of truth: packages/cli/kiri_cli/main.py and the per-group modules. Listed in alphabetical order.

Top-level commands without a group: init, plan, apply, doctor, onboard, use, targets, adapters, diff, lint, verify, explain, eject, discover, fetch, optimize, vacuum, history, reload, design, timetravel, map-source, graduate, upgrade.

The table above is not exhaustive — run kiri --help for the authoritative list of every group and command in the version you have installed.

Also installed by this package: kiri-mcp (the MCP server — see Use Kiri via MCP from your IDE) and kiri-job (a scheduler-safe entry point used by compiled platform jobs).

Conventions

• Commands resolve a target profile from kiri.yml via --target / -t and a project directory via --project / -p.
• Every long-running command writes a row to the apply-log — see _start_apply_log / _finish_apply_log in kiri_cli.main.
• Output defaults to JSON for machine consumers; pass --format markdown (or --format text) where supported for a human-readable rendering.
• AI calls go through kiri_ai_gateway; never directly to an SDK. Every call is audit-logged with prompt, response, model, cost, caller.

Documentation

• Command reference + guides: https://kirimana.io/docs
• CLI quickstart: https://kirimana.io/docs/getting-started/cli-quickstart
• Install + deploy: https://kirimana.io/docs/install/databricks
• Architecture principles: https://kirimana.io/why

Offline, with no network and no browser — the CLI documents itself:

kiri --help                  # every command group
kiri <group> --help          # drill into one
kiri onboard                 # write llms.txt: brief any LLM on the whole CLI
kiri onboard --agent claude  # or a CLAUDE.md for Claude Code
kiri doctor                  # ranked, fix-hinted health punch-list

License

FSL-1.1-ALv2 — source-available; each release converts to Apache-2.0 two years after it ships. The full text ships inside this package (kiri_cli/LICENSE), and the ODCS spec sliver stays Apache-2.0 forever (kiri_odcs/LICENSE).

Support

Kirimana is in beta and the repository is invitation-gated. For help, bug reports, or feature requests: https://kirimana.io/contact.