Proactive assistant

What this is for. The proactive assistant watches for work that needs you: overdue conversations, stale threads, anomalies, opportunities, reminders. It does not replace the Board or Home. Home’s Attention tile can show the same alerts; this page is the full list plus Learned Lessons. Keep the heartbeat off until you understand approval and spend — it makes paid model calls on a schedule.

Route: /proactive. Title: Proactive Dashboard. Subtitle: Active alerts, insights, and learned patterns. Also surfaces on Dashboard Needs attention → Alert items.

When to use it

You want EYAS to nudge you when work is overdue or a conversation has gone stale.
You enabled Proactive heartbeat under Autonomy and need the operator surface for what it found.
You want a one-shot Check Now instead of waiting for the next heartbeat.
You are reviewing lessons the assistant learned from earlier alerts.

Typical workflow

Enable Proactive heartbeat under Autonomy (Settings → Autonomy & self-improvement) only if you want background spend.
Open Proactive in the sidebar (/proactive).
Read Active Alerts. Priority badges: Urgent / High / Normal / Low. Types: anomaly, opportunity, reminder, insight.
Press Check Now to run an evaluation immediately. Empty state: All clear — no active alerts.
Scroll to Learned Lessons for patterns the assistant already applied (confidence %).

Features

When enabled under Autonomy, EYAS periodically evaluates whether to notify you or act (within policy). Alerts appear here and as Home Attention items.

The briefing text is written by EYAS’s background model in one isolated call: the Heartbeat routing tier (primary, then fallback), then the install default, then API providers, then CLIs that can run isolated calls — never a provider the gateway picks on its own, and never a CLI running with its own tools and memory. When no model qualifies (for example a Grok-only or Kimi-only install before their isolation is verified) or the budget is at stop, EYAS makes no model call and sends the canned alert Heartbeat: items may need your attention with the list of reasons. See Routing & budget — The background model.

Background runs get the same memory as chat. When the heartbeat or a board card starts a background run, that run receives the same recalled-memory block, with the current date and time, as a chat turn (see Memory — How recall reaches the model). A card’s goal is also remembered — once per distinct goal, however often the run is retried — as text EYAS wrote, not yours.

One runner for background runs. Cards in bot-listen or auto-assignee stages, and assign_task cards, run with the same setup as a retry of the same card, a scheduled agent routine and a colleague started by a hand-off: supervised, autonomous and gated by the autonomy ladder, on the card’s model, with memory recall, attached designs, documents, durable-memory capture and the completeness critic. Board runs now also get designs, documents and durable-memory capture, which they lacked before. A background card always takes its provider and model from one binding, never a card’s provider combined with an agent’s model from another provider.

The heartbeat can emit SLA breach signals (slaBreaches) when work drifts.

Signal kind	Typical meaning
Overdue	Conversation / activity past due date
Stale	Conversation idle too long while still open / working

Treat these as operator attention surfaces — combine with Board priority and Home setup recommendations.

Fields and controls

Active Alerts

Control	Meaning
Check Now	POST `/proactive/check` — run an evaluation now
N urgent	Count of urgent + high alerts
Priority badge	Urgent / High / Normal / Low
Type	anomaly · opportunity · reminder · insight
Title / body	Alert copy
Optional action button	Label from the alert (`actionLabel`) — opens the related URL when present
Timestamp	When the alert was created

Learned Lessons

Field	Meaning
Title / summary	Lesson copy
N% confidence	How sure the assistant is
Applied at	When it was applied, if present

Empty: No lessons learned yet.