AI development console

Experimental development surface

The console, analysis pipelines, and provider integrations are outside the stable backend contract. They are not required for REST, synchronous MCP, or Durable Operations. See stability labels.

The console accepts development questions and returns analysis or suggestions. Some paths use local rules and project inspection; others call a configured provider. Results depend on loaded application context and provider behavior. The console is not an authenticated CRUD API for application users or a durable workflow service.

Verify the CLI entry point

aksara ai flows chat "hello" --format json

This greeting is handled locally without a provider. The JSON result has ok: true, intent: "greet", flow_type: "conversational", and execution.mode: "conversational". It proves neither provider connectivity nor correct analysis of your models.

For the available syntax:

aksara ai flows chat --help

--provider and --model select overrides for provider-backed requests; --format accepts text or json. The top-level aksara ai chat path in older examples is not registered in 0.7.0.

Ask about application context

In a configured development application, prompts can ask to explain a model, review a route, suggest indexes, or investigate a problem. The implementation routes through conversational handling, investigation or intent analysis, and flow/provider paths as appropriate. It is not a fixed list of eleven intents, and not every prompt produces the same response fields.

Inspect ok, error, and error_code before using a result. execution can contain a local report or a provider response; prompt_pack can be absent or null. Treat these experimental shapes as development output, not a stable integration contract. Provider availability and result quality require separate testing with your application.

Studio access

When Studio is enabled and accessible under your configured authentication, its console uses POST /studio/ai/console with a message and optional provider_override and model_override. Suggestions use GET /studio/ai/console/suggest?q=.... These are Studio development endpoints; do not expose them as a public application tool boundary. Follow Studio security guidance before enabling access.

Autocomplete uses local intent suggestions. A real analysis request can include application metadata in provider context. Review that context and your provider configuration before sending development information externally.

Use results deliberately

Console output is analysis and suggested next actions. Review code changes, SQL, and migration plans using the ordinary application development process. Experimental ai plan apply is a separate mutation-capable interface, not an implicit consequence of receiving a console answer.

For agents that need to execute real application actions, use the authenticated MCP tutorial. For work that must survive retries or permission changes, use Durable Operations. Neither guarantee is established by a successful console response.