Dashboard
Your operational overview
Jobs
Active background tasks — auto-refreshes every 30s
| Task | Session | Status | Events |
|---|
Projects
Browse and open project workspaces
Profile
Your identity on Trinity
About You
This information is visible to admins and helps identify you on the platform.
Visibility
Account
Password
Session
Sign out of Trinity on this device.
Injected as context at the start of every chat. Use this to tell Trinity about your role, preferences, or recurring constraints — like ChatGPT's custom instructions.
Connect the facilities you use — each button grants just that facility's access. You can also pick a facility right at sign-in.
↓ Show token details
OLCF uses project access tokens that expire and must be renewed manually from myOLCF ↗ → Projects → API Tokens.
▸ ALCF Projects & Allocations
▸ NERSC Projects & Allocations
▸ OLCF Projects & Allocations
Queue, allocation & username are saved per system — switch the Default system to view/edit each. Jobs use the values for whatever system they target.
Advanced / troubleshooting
Token status & OLCF token entry moved to Authentication. These are fallback/debug tools for when the redirect login fails.
↓ Paste Globus token manually
Per-user environment variables injected into every Claude session. Use this to set service credentials like ClearML, GitHub, or API keys that override shared server defaults. Values are stored encrypted per-user and never visible to other users.
Click to prefill the key name above.
Download all your conversations. Each session includes messages, timestamps, and token usage.
Trinity is an agentic platform for autonomous scientific computing on DOE leadership systems. Describe your science in plain language and Trinity plans and runs the whole workflow — choosing a facility, staging data, writing and submitting jobs, monitoring them, and returning results — across ALCF (Polaris, Aurora, Crux, Sophia), NERSC (Perlmutter), and OLCF (Frontier). It is built on the Claude Agent SDK and rides Globus for identity, data movement, and compute, so a single conversation can reach every facility you are authorized to use.
Under the hood it composes reusable building blocks — Agents (domain specialists and multi-agent campaigns), Workflows (curated HPC pipelines), and Skills (atomic capabilities) — and records its work as Knowledge, Traces, and Provenance so results stay auditable and reproducible. Trinity is multi-tenant: each user's tokens, sessions, and settings are isolated. Built and maintained at the Argonne Leadership Computing Facility.
Rooms
AgentHub
Select agents and compose a multi-agent science campaign.
WorkflowHub
Loading…HPC workflow catalog — click Use to start a session with this workflow.
SkillHub
Loading…Atomic skills & playbooks — click Use to invoke one in chat (e.g. /submit-job).
Knowledge Store
Loading…Validated scientific K-atoms — confirmed domain facts used by agents during campaigns.
Campaign Provenance
Admin onlyFull audit trail of autonomous campaign decisions — traces, tool calls, alternatives considered.
Admin
Users
| Username | Role | Projects | File Write |
|---|
AgentTrack
Interactions with other agents across all your sessions.
Collaborators
Connect with people, then start a group chat. Both sides must accept.
Usage
API tokens & cost across your sessions.
Tests
Trinity systematic test-suite — does it work & what does it cost.
Trace Timeline
Select a session and click View Trace
Eval Lab
0 runsModel Playground
Loading…