Announcements

Dev update: the v0.9.0 release plan, revision 2 (plan PR #2096). Freeze Wednesday Aug 19. Headline: Workspace ships, skills library ships safely, schedules work again — what lands now, what blocks the cut, and what rides if ready.

Thursday, July 30, 12 PM ET / 9 AM PT, live on Zoom, free. Converting an entire fleet of existing agents — built with Claude Code, n8n, LangChain, or any other stack — into a validated, Trinity-compatible agent fleet, step by step. Hosted by Eugene Vyborov, creator of Trinity.

The 5 core rules for managing agent knowledge and a production-ready pattern that separates canonical truth from operational data: the Decisions / Facts / Judgment framework, Knowledge as Code with Git, a central "mind center" agent, and why human endorsement gates the canon.

773 commits, ~95 issues. Task-completion events, agent self-reminders, shared rooms, skills as full packages, display names, ghost agents, bring-your-own GitHub token — plus voice replies v2, local-first telemetry, a fleet-hygiene pass, and a blast-radius guard on data retention.

Thursday, July 23, 12 PM ET / 9 AM PT, live on Zoom, free. Building a company knowledge base with Cornelius on Trinity — from empty session to a deployed agent, built live. Hosted by Eugene Vyborov, creator of Trinity.

An animated walkthrough of how Trinity works — one scheduled task followed end to end: the agent container, the admission gate, delegation between agents, human approval, the audit ledger, circuit breakers, capacity queues, channel adapters (Slack/Telegram/WhatsApp), and the network security model.

Five feedback loops, no retraining: the paper behind Brier HQ — a forecasting fleet running on Trinity since February 2026 — is now on GitHub, with a video walkthrough. 96,161 predictions, 73,552 resolved, Brier 0.194 vs 0.250 baseline.

The end-to-end pattern for deploying sovereign, isolated AI agents for each of your clients on open-source Trinity: one agent per client, full data and credential isolation, live agent onboarding with Claude Code plugins, a central Cornelius knowledge base, client channels on Slack/Telegram/WhatsApp, and real guardrails.

110+ issues across the platform: voice replies on Telegram, Slack & WhatsApp, the Brain Orb live 3D knowledge map, the Grid dashboard, enterprise identity (2FA, SSO, permissions matrix), loop guardrails, webhook signing, and agents-as-MCP-tools. Release tour video included.

PostgreSQL is now the recommended production database (SQLite stays the zero-config dev default; end-of-support 2026-09-01). Plus the OpenAI Codex harness MVP, agent runtime data volumes, credential hot-reload rotation, fire-and-forget execution dispatch, and TOTP two-factor auth. 36 issues shipped.

From an empty Claude Code session to a deployed, autonomous recruiter agent — live, ~38 minutes. "Scout" interviews candidates over Telegram, scores them against a rubric, and keeps a human in the loop for the hiring call.

34 issues: sequential agent loops shipped end-to-end with a web UI, agents place outbound phone calls (VoIP Phase 1, beta), Health/Ops/Executions merged into one Operations console, push-heartbeats, and secure-by-default channel access.

~35 minutes on loop engineering: the 4-step playbook pattern (read → work → verify → write), keeping loops safe with max runs, stop signals, timeouts, and cost ceilings, loops vs. schedules, and why remote loops on Trinity beat laptop loops.

The four layers of a production loop (termination, progress detection, verification, durable state), how loops fail in the wild, Trinity's loop primitive with hard run caps and evidence-terminated stops, and remote loops that let you disconnect.

Agent coordination moves from push to pull / work-stealing: every task is a durable row in a per-agent Postgres queue, and agents pull work only when they have a free worker.

Hands-on workshop session: build a personal medical advisor agent and deploy it to Trinity end-to-end.

A place to ask questions, propose features, and show what you're building — Q&A, Ideas, Show and tell, General, and Announcements categories.

264 commits, 117 issues closed, 10 contributors. Stateful Session tab, experimental Voice Workspace, mandatory Redis auth, non-root containers, and A2A Agent Cards.

Weekly Trinity workshop on Cornelius — a Zettelkasten knowledge-base agent combining vector and graph search for advanced knowledge management.

Trinity as a platform — sandboxing, governance, audit trails, credential injection, multi-channel access — plus Cornelius's brain dependency graph, hybrid retrieval, and perception/deliberation loops.

Three new autonomous capabilities for the knowledge base agent: an overnight incubation loop, domain watch, Socratic insight interviews, plus YouTube transcript extraction.

Weekly workshop walking through running the Cornelius knowledge base agent on Trinity.

A full platform walkthrough covering most of the recent features in one demo — a good recap if you've been following development.

Cornelius — 3,000+ knowledge nodes — now talks back in real time using Gemini voice. Demo shows voice mode for any agent in Trinity.

abilityai/trinity-ops-public is a Claude Code agent for operating your Trinity instance — health checks, restarts, updates, rollbacks, and diagnostics from one .env.

Biggest release yet: WhatsApp per-agent binding, Outbound File Sharing, Conditional Scheduling, Voice Tool Calling, Platform Audit Log, Reliable WebSocket Delivery, and more.

Day-to-day work merges into `dev`. `main` is release-only. Feature branches, PR targets, and the new /release skill explained.

Walkthrough of the 3 agents running the Trinity platform in production — Dev, Ops, and Docs agents. 160,000+ LOC, 3,000+ tests.

Workshop #3 recording is up; Workshop #4 starts in 45 minutes — agents that build Trinity, running inside our own infrastructure.

docs.ability.ai launch, Ask Trinity Q&A agent, Redis Streams event bus, OpenTelemetry tracing, Guardrails Phase 1, and more.

Today's session explores how agents talk to each other — architectural patterns and hands-on collaboration experiments.

Every Tue & Thu at 8:30 AM PT. 45-60 min on Zoom. Build autonomous agents that delegate, schedule, and collaborate.