Skip to main content

​
Real-time streaming

MIRA renders responses as they arrive — token by token for NAE, iteration by iteration for RLM. You see the answer being constructed rather than waiting for it to complete. NAE streaming: Each token from the LLM API arrives via the nae:chunk IPC event and is appended to the message bubble in real time. Rendering is throttled to ~60 fps using a requestAnimationFrame flush loop — fast model responses don’t block the UI. RLM streaming: Each reasoning iteration (code written, code executed, output observed) arrives as an rlm:iteration event. The REPL Console shows the full iteration stream. The final answer arrives as an rlm:final event and is rendered as a complete Markdown block.

​
Markdown rendering

All assistant responses are rendered as Markdown with these elements supported:

​
Code blocks with syntax highlighting

Code blocks are syntax-highlighted based on the language tag:
Supported languages include Python, TypeScript, JavaScript, JSON, SQL, Bash, YAML, and dozens more. Unrecognised language tags render as plain monospaced text. Each code block has a Copy button in the top-right corner.

​
Copy message

Every message bubble has a Copy button (clipboard icon) that copies the raw Markdown source to the clipboard — not the rendered HTML. This preserves formatting when pasting into other tools.

​
Timestamps

Messages show a relative timestamp (e.g. “just now”, “2 min ago”) that updates in real time. Hover over a timestamp to see the absolute time.

​
Token usage

When a response completes, NAE reports input and output token counts in small text below the message. This helps you track context usage across a long session.

​
User message editing

User messages are displayed as plain text. They cannot be edited after sending — create a new message to refine your query.