Trinity v0.6.1 — Agent Loops End-to-End, VoIP Calls, Unified Operations

Trinity v0.6.1 is out — 34 issues: agent loops end-to-end, VoIP calls, and a unified Operations console.

See the full write-up in What's New →

Agent loops shipped end-to-end

The run_agent_loopMCP tool runs bounded sequential task loops — max runs, stop signals, per-run cost and timeout — now with a web UI on Agent Detail and a distinct Loops bucket in timeline views. Start a loop, disconnect, review the run-by-run trail later.

Agents can place phone calls

VoIP Phase 1: outbound calls over Twilio Media Streams + Gemini Live, with transcripts persisted and handed back to the agent after the call. Off by default, bring-your-own-Twilio.

One Operations console

Health, Ops, and Executions merged into a single tabbed nav area with a unified badge and per-tab Clear All. Agent Detail now lands on an Overview dashboard — execution trends, health, recent activity — and agent config moved into a sectioned Settings tab.

Reliability + security

• Agent push-heartbeats for fast failure detection; fleet health monitoring survives backend restarts
• Fleet-wide executions dashboard + per-schedule analytics
• New agents now require verified email for channel access by default (owner-overridable)
• trinity-cli 0.2.5 is on PyPI with the embedded credential removed — update if you use the CLI

Model catalog refreshed: Claude Opus 4.8 added, retired Gemini models swept out of voice, transcription, and image flows.

Full notes (all 34 issues) → github.com/Abilityai/trinity/blob/main/docs/releases/v0.6.1.md