Agent Intelligence Dashboard, Alert Deduplication, and CLI Reliability Fixes
This release ships a new Agent Intelligence dashboard, centralizes all agent-to-founder communications through a single email path, collapses redundant swarm alerts into single records, and fixes a set of CLI reliability issues found during deep testing.
Agent Intelligence Daily Digest
A new agent-intelligence system now generates a daily digest of behavioral insights for each agent — pulling from eval history, score trends, heartbeat patterns, and anomaly signals. This information is exposed at /dashboard/agent-intelligence and via a new on-demand API endpoint, giving operators a consolidated view of how their agents are evolving over time.
The daily cron runs at 2am UTC and produces per-agent summaries that surface: score momentum (improving / stable / declining), top-performing eval dimensions, failure pattern frequency, and recommended next evaluations based on gap analysis.
Centralized Founder-Mailer
All agent-generated emails now route through a single emailFounder() function with built-in rate limiting (4 emails per agent per 4-hour window for normal priority, always-send for high/critical). This prevents inbox flooding from talkative agents while ensuring high-urgency signals always arrive immediately.
Previously, different agent loops used different email paths with inconsistent formatting and no rate control. The new system applies a consistent template with role-colored headers, urgency badges, and structured data tables regardless of which agent sends the message.
Alert Deduplication
The user_inbox table now deduplicates repeated alerts using a composite key of (from_agent_role, alert_type, dedup_key). When the same alert fires multiple times, subsequent occurrences increment a counter rather than creating new rows. This eliminates the noise from frequent cron-driven status alerts that previously created dozens of identical inbox entries per day.
CLI Bug Fixes
Five reliability issues were fixed in the Armalo CLI following a comprehensive test run:
room emitnow correctly validates event types before sending — previously accepted invalid entry types that the server rejectedroom intervenefixed the same entry type validation gap- Error messages on failed API calls now include the HTTP status code for faster diagnosis
agent listpagination now handles orgs with 0 agents without returning a parsing erroreval runno longer silently swallows timeout errors — they now surface with a descriptive message and exit code 1
OpenAI Multi-Provider Support in Admin Swarm
The admin swarm now supports OpenAI models alongside Anthropic as a second inference provider. The OAuth Max plan credential path handles both, with automatic provider selection based on task type. This enables fallback behavior when primary provider capacity is constrained.
All fixes ship to production immediately via Vercel. CLI updates require npm install -g @armalo/cli to pull the latest version.