AI development commands¶

Experimental

These development tools sit outside the stable backend contract. Analysis, provider quality, planning, and Studio internals remain experimental. They are not required for REST, synchronous MCP, or Durable Operations. See stability boundaries.

The installed CLI has several distinct AI command families. Use actual subcommand help and the generated reference to find their arguments. Historical ai query, ai generate, ai doctor, ai agent, ai patch, ai ask, and ai config commands are not registered in 0.7.0. A code-generation or autonomous-execution example using those names does not describe a supported shortcut.

Discover the available surface¶

aksara ai --help
aksara ai flows --help
aksara ai run --help
aksara ai plan --help
aksara ai-provider --help
aksara ai-hub --help
aksara agent --help

Family	Purpose
`ai context`, `ai hints`	Assemble development context or inspect hint metadata
`ai schema-health`, `ai schema-issues`	Inspect schema analysis reports
`ai flows`	Flow-oriented analysis and console entry points
`ai run`	Provider-backed flow execution
`ai plan`	Generate a JSON plan template, preview a file, or apply a plan
`ai-provider`, `ai-hub`	Provider configuration and inspection
`agent`	Development context, prompts, playbooks and workflow analysis

The names are not interchangeable. In particular, experimental plan application is not a Durable Operation, and an analysis suggestion is not an approval grant.

Inspect schema guidance¶

With your models and database configured:

aksara ai schema-health --format json
aksara ai schema-issues --severity danger --format json

See Schema Doctor for prerequisites and report meaning. These commands do not automatically fix models or apply migrations. For deployment readiness, use Doctor instead.

Try the console without a provider¶

aksara ai flows chat "hello" --format json

The greeting path returns a local conversational response. This verifies the console entry point, not model quality or application analysis. Other prompts can require a configured provider and can send assembled context to it; inspect your provider and application configuration before running those prompts.

See the console guide for its boundaries. The correct CLI path is ai flows chat, even where historical help examples omit flows.

Work with a plan file¶

aksara ai plan template --intent "Describe the ticket model" --mode read
aksara ai plan preview --help
aksara ai plan apply --help

template prints a starter JSON structure. It does not infer a complete plan from the description. preview takes a plan file or stdin; apply is a separate mutation-capable command. Neither is the free-form ai plan "do this" interface shown in older documentation. Review plan operations, configuration, and code changes independently before applying one. This release does not claim an end-to-end provider or plan-application validation from help output.

Configuration and safe MCP use¶

Use the settings reference and provider guide for configuration. Do not pass actual provider keys in copied command examples or attach them to diagnostic evidence.

To expose application actions safely to an external agent, follow the official MCP client tutorial. Development console access is not a substitute for that authenticated, scoped application execution boundary.