Use Vera from scripts

Use the command line for one-off prompts and exports. Use vera stdio to control a hosted conversation through JSON messages. For application code, see the embedded SDK or Python workflows.

Send a single prompt

Print mode sends a prompt from the terminal. Add --prompt-only for a request containing only Vera’s identity prompt and your message, with no tools:

vera -p "Summarize this text" --prompt-only

Choose the model and effort for one run

--model selects a model from your library for this turn. --effort overrides its reasoning effort for this turn; the selected model must support that level. If --effort is omitted, no per-run effort override is applied.

vera -p "Count the pirate crow's feathers." --model "provider/model" --effort low

Replace provider/model with a model ID from vera library list.

Choose a minimal startup

Both also work without print mode. The selected startup mode is saved with the conversation and restored on resume.

Exchange JSON messages

vera stdio connects to the resident host and exchanges one JSON object per line. Start a new conversation, attach to a live one, or resume saved work:

vera stdio
vera stdio --attach <agent-id>
vera stdio --resume <session-id|path>

The exchange includes prompts and responses to UI requests. This is the message shape, with omitted fields shown as ellipses:

{"type":"prompt","content":"Review this file"}
{"type":"ui_request",...}
{"type":"ui_response","requestId":"...","response":{...}}

Export a conversation

Export the active conversation branch as Markdown:

vera export ~/.vera/runtime/sessions/<session-id>.jsonl

Add --format json for the same user-facing transcript as versioned JSON. Output goes to stdout; redirect it to save a file. Export does not modify the conversation or workspace.

Inspect a recorded request

vera inspect ~/.vera/runtime/sessions/<session-id>.jsonl

This prints the latest recorded provider-neutral request: system prompt, model-visible messages, and offered tools. It is not the provider’s wire payload.