Skip to main content

About TritonAI

TritonAI Roadmap

A public, status-based view of approved TritonAI work from the 2023 chatbot launch through the 2026 governed builder path. The platform now serves 73,000+ students, faculty, and staff. Completed work links to verified program updates when a corresponding public update is available.

How to read this page

What each status means

  • ShippedPublicly available
  • PilotBounded testing with oversight
  • In developmentActive work without a committed launch date
  • ExploringDiscovery only; no delivery commitment

Current horizon

2026 delivery detail

Each quarter groups supervised solutions, the reusable pieces underneath them, and the work of getting prototypes to supported services. A quarter's status covers the group as a whole, so an individual service linked below may be further along.

Q1 2026 — Agents

Pilot

Ecosystem Expansion

Launched transcript OCR and student record matching, expanded instructional AI to 8 courses with scheduling assistant pilots, and opened the TritonAI site with API access and early N8N workflows.

  • Transcript verification moved from an OCR and matching proof of concept toward an OnBase handoff for enrollment operations.
  • Instructional AI added one-click assistant creation while scheduling and spring-course pilots continued.
  • Contract review expanded as a supervised Procurement workflow with configurable, policy-backed redlining.
  • The TritonAI site and Developer APIs opened governed model access for campus builders.
  • Research proposal risk and research alignment proofs of concept tested focused research-support workflows.

Owner TritonAI solutions team and sponsoring unitsReviewed 2026-07-27

Q2 2026 — Refine

Pilot

Process Automation

The builder path expanded, and TritonGPT moved into the campus mobile app. A Canvas-integrated instructional assistant shipped. Teams also advanced pilots in finance, transcript processing, and contract review. Research, BioBib, and observability pilots moved ahead as well.

  • Developer API access and N8N Agent Builder broadened the path from model access to supervised workflow orchestration.
  • TritonGPT extended into the campus mobile experience and began surfacing live campus signals with stronger response traceability.
  • A Canvas-integrated instructional assistant established a reusable delivery pattern for course support.
  • Transcript matching reached production with an OnBase handoff for enrollment while contract review graduated to a supervised Production workflow.
  • BioBib formatting, risk-based proposal review, and AI IT observability expanded the bounded prototype portfolio.

Owner TritonAI solutions teamReviewed 2026-08-10

Q2–Q3 2026 — Expand

In development

Multi-Institution Expansion

The platform is expanding beyond UC San Diego as a governed multi-tenant service shared across UC and partner institutions.

  • BearGPT is live at UC Berkeley as a partner deployment of the shared TritonAI gateway and governance model.
  • ANR AI serves staff at UC Agriculture and Natural Resources under a renewed multi-year services agreement.
  • FredGPT is live at Fairleigh Dickinson University as a research-oriented institution instance.
  • UC Merced is actively deploying its instance of the shared platform.
  • Partner instances inherit the same hosting, model choice, trust, and observability controls as the UC San Diego deployment.
  • The expansion reuses the existing platform rather than duplicating governance and infrastructure per campus.

Owner TritonAI platform teamReviewed 2026-08-10

Q3 2026 — Ecosystem

In development

Governed Builder Path

Launching the TritonAI Harness pilot and initial hosted applications, organizing reusable skills and connector patterns, and preparing governance and observability for broader agentic development.

  • A bounded Citizen Developer pilot is testing a gateway-connected harness, tiered model access, and early project accountability patterns.
  • The TritonGPT Class Planner is moving through a supervised student rollout tied to registration readiness.
  • Campus App Hosting is beginning with limited-scope proof-of-concept apps and a shared-responsibility path to support.
  • The skills pipeline is organizing reusable UC San Diego skills, MCP integrations, and data-source patterns.
  • Shared knowledge work is testing reviewed promotion, source provenance, freshness checks, and scoped retrieval before context can be reused.
  • Academic advising remains in a scoped prototype and build phase.
  • Governance and observability work is documenting data, security, reliability, and operational requirements before broader access.

Owner TritonAI platform and solutions teamsReviewed 2026-08-10

Beyond Q3 2026

Exploring

Governed agency at campus scale

Evaluating repeatable, supervised agent workflows that can use approved tools and connectors while retaining accountable human review.

  • Agents would retrieve approved data and call approved tools only within bounded workflows.
  • Shared context would move from private task context to reviewed team or service knowledge through explicit promotion.
  • Human review would remain at consequential decisions and actions.
  • Observability would capture the reliability, usage, and evidence needed to operate services responsibly.
  • Successful pilots would graduate through a named service owner, governance review, support model, and approved hosting path.

Owner TritonAI program and campus governance partnersReviewed 2026-07-27

Foundation

How we got here

Earlier milestones, from opening up broad access to building the shared infrastructure and the first focused campus workflows.

2023 — Alpha

Shipped

Beta & Training

Secured Chancellor's Office funding, invited 400+ staff to the campus beta, and launched AI Essentials I in the UC Learning Center.

Owner TritonAI program

Q2 2024 — Live

Shipped

Campus-Wide

Opened campus-wide availability for ~37K employees, upgraded to Llama 3 with campus SSO, and refreshed the UI with file upload.

Owner TritonAI platform team

Q3–Q4 2024 — Growth

Shipped

Adoption

Grew Fund Manager Coach adoption, expanded departmental use cases, and began infrastructure planning for the next platform phase.

Owner TritonAI solutions team

Q1 2025 — Build

Shipped

Platform Prep

Upgraded the platform architecture for scale and prepared the model gateway for both commercial and local LLMs.

Owner TritonAI platform team

Q3 2025 — Integrate

Shipped

Widget & Infra

Deployed the Blink AI widget without a login requirement and integrated LiteLLM to enable model choice.

Owner TritonAI platform team