What's New in Trinity v0.6.1
Released 2026-06-12
A UI/UX-and-reliability release. The biggest change is how you watch your fleet: Health, Ops, and Executions are now one Operations area, and every agent opens to a new Overview dashboard. Sequential agent loops ship end-to-end (start one from the UI or from Claude Code), agents can now place phone calls over voice (beta), new agents are secure by default, and the model list adds Opus 4.8.
Improvements
•One Operations area — Health, the operator queue, and Executions are consolidated into a single tabbed Operations page, with a unified notification badge and a per-tab Clear All. Old
/monitoring, /executions, and /operating-room links redirect here automatically.•Agent Overview dashboard — Each agent now opens to an Overview tab: activity trends, success rate, context use, and recent activity at a glance. Tabs that don't fit collapse into a More menu instead of scrolling off-screen.
•Sequential agent loops — Run the same task against an agent repeatedly with a bounded run count and an optional stop signal, from the new Loops tab or via MCP / Claude Code.
•Agents can place phone calls (beta) — An agent can dial out and hold a live voice conversation over the phone. Off by default; an admin enables it.
•Per-schedule and per-playbook analytics — Duration, token, and tool-call distributions for each schedule and playbook.
•Read the operator queue from MCP — New tools let agents and scripts read pending operator-queue items, broadly or by agent name.
•Faster failure detection — Agents push a lightweight heartbeat, so the platform notices an unresponsive agent quickly.
•Secure-by-default access — Newly created agents default to "require verified email." Existing agents keep their current policy; admins set the fleet-wide default in Settings → Agent Defaults.
•Settings tab — The per-agent Guardrails tab is now a sectioned Settings home for agent configuration.
•Refreshed model list — Adds Opus 4.8 and removes end-of-life models across the app.
•Loops shown distinctly — Timeline and execution charts display loop runs as their own category instead of folding them into Scheduled.
Fixes
•No more false "0 tool calls" failures — Long tasks whose child processes held the output pipe are no longer mislabeled as empty failures.
•Subscription auto-switch no longer kills running work — Switching a rate-limited subscription drains in-flight executions instead of dropping them.
•Phone and voice connect reliably — Fixed a call-answer failure and a "model is required" voice error on Docker-compose deploys.
•Fleet health survives restarts — The health-monitoring loop now resumes automatically after a backend restart instead of staying off.
•Accurate timeout messages — A run stopped by the watchdog is no longer mislabeled "timed out."
•More build/scratch space for agents — Agents no longer run out of temporary space during pip, npm, or build steps.
•Reliable resource configuration — CPU/memory display, dialog pre-selection, and the apply flow are fixed.
•Telegram voice and image generation work again — Removed a retired model that was returning 404s.
•Accurate notification count — The navigation badge shows the real pending count instead of capping at 1.
•Approval notifications — A requester is now notified when their channel-access request is approved.
•Correct version display — The version endpoint shows the real build version instead of "unknown."
•Operator sessions protected — The orphan sweeper no longer kills active operator terminal sessions.
Security & Hardening
•Unified, consistent request rate limiting across endpoints.
•Removed an embedded credential from the
trinity init CLI command and published a patched CLI.•Resolved critical and high-severity dependency advisories.
Upgrade Notes
•Re-login required after the platform restarts, and MCP clients need to reconnect.
•New agents default to "require verified email." Existing agents are unchanged; change the default in Settings → Agent Defaults.
•Per-task timeout override removed. Task dispatch now uses the agent's execution-timeout setting; schedule-level timeouts are unaffected.
•Update your bookmarks.
/monitoring, /executions, /operating-room, and /events now redirect to /operations.•Self-hosted operators: no special migration beyond the standard upgrade — agents pick up the new temp-directory default on their next restart.