yana-ai

yana-ai [command] [target] [flags]yana-ai help | -h | --helpyana-ai version | -v | --version

yana-ai is the Python CLI for Yana AI, a local control plane between a coding model and real power. A model proposes an action; Yana checks authority, capability, and policy (including a human gate) before anything runs. The installed command is yana-ai (`pip install yana-ai`). The script is also named yana; the PyPI entry point is yana-ai.The wrapper (`bin/yana`) dispatches to Python helpers under `core/scripts/` or to the native runtime yana-rt. Commands such as audit, chat, task, graph, and hunt need yana-rt on PATH (install with `cargo install yana-rt`). verify, init, policy, guard, install, giamthi, router, watch, and several report commands run in Python without it.install writes `.yana-aiignore`, gitignore entries, an MCP template, an example GitHub Actions workflow, and harness files for Claude Code and/or Codex. doctor checks Python, git, ignore rules, Claude/MCP config, API keys, and whether Yana hooks are wired. audit is yana-rt scan: it scores the agent setup and can emit Markdown or SARIF. giamthi installs an opt-in OS supervisor (systemd timer on Linux, launchd on macOS, Task Scheduler on Windows) that compares enforcement files against a human-approved baseline every 6 hours and writes `GIAMTHI_HALT.lock` on drift; a hook then denies later tool calls until a person deletes the lock.Yana Studio, Yana Desktop, and Yana Web are separate GUI apps. This command is the shell interface. The product is not published to npm. Apache-2.0.

--json

Machine-readable output on doctor, audit, map, score, hunt, and other read commands

With audit / hunt, exit non-zero on findings at level or above (`low`, `medium`, `high`, `critical`)

With audit, scan only files changed since base (for example `origin/main`)

With install, which harness files to write. Default all

With install, giamthi, fix, and others: print what would change

With chat: `anthropic`, `openai`, `kimi`, `ollama`, or another catalog id

Print the built-in command list

Print `yana-ai` and the package version

.yana-ai/config.yml

Per-project CLI settings. yana-ai config show / set / init

Finding IDs to suppress, gitignore syntax. install creates a default file

Daily and monthly spend ceilings for cost set-policy / cost guard

Capability leases granted with lease grant

Repo-supplied integration config. Unused until a person runs yana-rt trust allow (not reachable through this wrapper)

install [target] [--engine all|claude|codex] [--guards] [--supervisor ask|install|skip] [--no-audit] [--dry-run]

One-command project setup. Default --engine is all (Claude + Codex). --guards also runs guard install all. --supervisor controls Giám Thị. Default target is `.`

Environment health. Exit 0 if healthy, 1 on warnings, 2 on failures. doctor dispatch cross-checks `src/main.rs` against this wrapper's dispatch table

Scan the AI-agent setup. Forwards to yana-rt scan. --fail-on is `low`, `medium`, `high`, or `critical`. --watch re-runs on file change. Reads `.yana-aiignore`

Interactive wizard (engine, profile, guards). The canonical implementation is Python; yana-rt init is a simpler flag-driven alternative

Check that safety hooks are wired and active

Interactive REPL through yana-rt. With no --provider, uses `ANTHROPICAPIKEY`, else `OPENAIAPIKEY`, else local Ollama. A missing cloud key exits; it does not silently switch providers. run_command still needs interactive approval

OS-level Giám Thị supervisor. run is a single pass. Uninstall keeps `GIAMTHI_HALT.lock` and audit evidence

Install runtime enforcement hooks into the project (Python). Distinct from yana-rt guard, which is the in-process PreToolUse fast path

Safe-config templates for audit findings. Needs python3

Task store in yana-rt. done requires --evidence

Check recorded task evidence

Blast-radius map: settings, MCP, workflows the agent can reach

Audit score and risk level

Secrets, code, dependency, and supply-chain scan

Knowledge graph over the target tree

Apply a known safe fix

Self-update the Python package to the latest release

Full command list, welcome banner, or version string

YANA_RT_BIN

Absolute path to a compiled yana-rt binary. Do not point this at the pip shim named `yana-rt`

Set by the wrapper to the installed package tree so `bin/yana` finds `core/`

Cloud providers for chat. Kimi uses MOONSHOT_API_KEY

VirusTotal hash lookup for filescan check. File contents are not uploaded

Opt-in Bash-output compaction in yana-rt compact (hook path, not this wrapper)

Needs Python 3.11 or newer, git, and a supported harness for the install/audit path. Many subcommands fail with exit 3 if yana-rt is missing. The pip command yana-rt is a wrapper that looks up a native binary; install the runtime with Cargo (or set YANA_RT_BIN). Destructive-command guards match command strings, not a full shell parse. Giám Thị is opt-in and can take up to 6 hours to notice drift. Cursor hooks screen shell commands, not MCP calls; Antigravity rules are advisory. Chat runs as its own process, outside Claude Code's PreToolUse hooks. Not distributed on npm.

Yana AI was started on 17 May 2026 by Vũ Văn Tâm in Vietnam (Apache-2.0). The Python package yana-ai is published on PyPI; the Rust runtime yana-rt is a separate crates.io crate. npm publishing of the product axis was dropped in July 2026.

yana-rt(1), claude(1), codex(1), aider(1), aichat(1), ollama(1), pip(1), cargo(1), git(1)

Source code · Homepage · Documentation