Skip to content

What's new

Dates below are when each feature reached the production Trinity server. Subscribe via the RSS feed to get new entries in your reader.

A large batch of updates reaches the production server:

  • Redesigned interface — a rebuilt, faster web UI is now the default, with a consistent visual language across chat, Settings, and Admin, and a theme-following look throughout.
  • Attach files in the composer — upload, drag-and-drop, or paste files straight into a message, the way you would in ChatGPT or Claude.
  • Artifact gallery — plots, reports, and other files your session produces are collected per session so you can find and open them in one place, with inline preview for images and video.
  • Better file viewer — LaTeX and Fortran syntax highlighting, optional one-click LaTeX compile, a line-wrap toggle, a collapsible file tree, and a manual refresh button.
  • Pick your model per chat — a per-session model picker, plus the ability to register your own model providers (with write-only API keys) from Settings.
  • Smarter sidebar — full-text search with ranked, highlighted snippets across folder names and conversation content; a By Project / By Recent toggle; activity ordering; and inline folder rename / move.
  • Sub-sessions — work that a chat spawns in the background now appears as its own nested session under its parent, and status dots show when it's your turn versus a background agent still working.
  • Interrupted chats resume — if a turn is cut off (a disconnect or restart), Trinity now safely picks it back up without re-running work it already did.
  • Reviews & scientist feedback — campaign owners can invite reviewers, and a review dashboard lets scientists read runs and write structured feedback back.
  • Modernized terminal — GPU-accelerated rendering, clickable links, in-terminal search, and automatic reconnect.
  • Monthly schedules — schedule recurring tasks on a day-of-month interval.
  • API access — mint a personal bearer token from Settings to call Trinity's /query endpoint from your own scripts. See API access.
  • Customizable keyboard shortcuts and a thinking-level control in the composer.

2026-07-29 — Trinity Hub launches

The community registry and this documentation site go live: register your own agent or skill and make it @mention-able in any Trinity room without running Trinity yourself. See Register an agent and Contribute a skill.

2026-07-11 — Usage tab, persistent sessions, opencode engine

  • Usage tab — track token usage and estimated cost by day, project, and model from the dashboard.
  • Persistent, steerable sessions — long-running chats survive reconnects, and you can steer a task while it's still running instead of waiting for it to finish.
  • opencode engine — Trinity can drive non-Claude models (including open-weight models on ALCF inference endpoints) through the same chat interface.

2026-06-29 — Three-hub model

Trinity's backend reorganizes around three hubs — Agent Hub, Skill Hub, and Workflow Hub — the foundation the community registry builds on. See How Trinity Hub works.

2026-06-19 — Group chat and mobile app

Shared rooms let multiple people and agents collaborate in one conversation; a companion mobile app (Expo) brings Trinity to iOS and Android.

Have a request or found a bug? Open an issue or reach out to the maintainer.