Technology · FlashBrain

The reflex layer.

Zaelar's own orchestration code — not a general-purpose reasoning agent. It answers every turn in under a second, drives the canvas, and decides what's worth escalating.

View full diagram →

Every input — a voice turn, a connector message, a UI click, a message from another agent on the network — passes an attention gate before it ever reaches a model: is this actually directed at the agent, or ambient noise? With the mic always open (no wake word needed), that call is judged by a fast model reading the sentence itself, not just a timer. Does it need a hard interrupt ("stop", "close everything")? That check stays a deterministic rule — it can never wait on a model. Only what's left goes to the prompt.

The same reflex loop drives a conversation with another agent over the agent network — but in a stricter profile: no tools, no memory of the operator, unless explicitly opted in.

One small prompt, composed live

The prompt isn't a static persona — it's assembled per turn from live state: identity and mission, what's on screen right now, a synthesized slice of recent conversation, plus a terse layer of available tools. A small, fast, non-reasoning model reads it and responds with either spoken text or a tool call. Which model that is ships as a sensible default and is yours to change from the config UI — never baked in.

Tool catalog

The model decides what to do by calling a function, not by matching keywords. That keeps the decision reliable even on a small, cheap model. The tools offered change with context — most only appear when they're actually relevant.

tool	what it does	when it's offered
escalate	Launches a Brain Worker	always
web_search	One factual answer, resolved in the turn (~1-2s)	always
recall	Reaches into durable long-term memory mid-turn	always
reveal_secret	Fetches an encrypted operator secret, out-of-band	when a vault exists
play_music	Fuzzy music requests, resolved in-turn	always
play_video	Loads a video in the youtube widget	when that widget exists
show_widget	Opens a widget by its name or an alias — never a guess	when widgets exist
show_panel	Opens the native side panel — chat, processes, schedules or networks	always
widget_data	Runs one declared action on an open widget	when widgets are open
manage_widget_alias	Adds or removes a name/alias a widget answers to	when widgets exist
reply_message	Drafts a reply, gated by a confirm step before sending	when a messaging channel is linked
delete_widget	Deletes a widget for good, with confirmation	when widgets are open
set_style_directive	Applies and persists a standing behavior rule	always
authenticate_web	Opens the browser to log in to a site	always
connect_cluster	Joins an agent network — private, or public with no credentials	always, behind a Yes/No confirm
cluster_send	Says something to a network, or to one agent in it	only when a network is connected
send_to_worker / stop_worker / answer_worker	Steers a live Brain Worker	only with live workers

Showing, hiding or moving something on the canvas is handled by lightweight text tags in the spoken response, not a tool call — the FlashBrain never touches the executor directly. Anything that needs real reasoning goes through escalate to a Brain Worker.