Skip to main content
Triton AI Triton AI

Get Started

  1. TritonAI
  2. Developer APIs
  3. Get Started

Build with TritonAI

Set up TritonAI Harness

Download and open the installer, get a TritonAI access key if you need one, then finish setup and try your first prompt.

Already opened the installer? Get your key. For another client or your own code, see other setup options.

Download and open the installer

Choose the stable installer for your computer. It guides you through access setup and installs TritonAI Harness. You can download it before requesting a key.

Primary Supported Client

TritonAI Harness

TritonAI Harness is the desktop workspace. TritonAI Installer sets it up on your computer and checks your access to campus models.

Windows (x64)

Requires an x64 processor. Run the downloaded setup file, then open TritonAI Installer. Windows may show a SmartScreen warning.

Download for Windowsx64 · Setup EXE

Read TritonAI Harness release notes.

When the installer asks for your TritonAI access key, use your existing key or follow the next step to request one. Installation completes after your key is checked.

Get your access key

Already have a TritonAI key? Continue to finish setup. If you need one, select Don’t have a key? in the installer to open the access guidance, or use the Kuali Build form below. If the installer sent you to this page, this is the next step.

  1. 1
    Submit the access form

    Describe your intended use and choose the access you need. Some requests require approval.

    Request Gateway access (Kuali Build)

  2. 2
    Retrieve your key

    The email contains a one-time link to view and copy your key. The key itself is not in the email. Keep the key private and save it securely before leaving that page.

  3. 3
    Return to the installer

    Return to TritonAI Installer with your key, then follow the next step to finish setup.

UC-hosted models carry no recharge for campus administrative work. Cloud models and sponsored research require a departmental or project chartstring. Recharge is the campus term for internal billing to a chartstring.

Eligibility and recharge details
  • Faculty and staff: Campus and Health Sciences personnel may use UC-hosted models for administrative work under monthly caps.
  • Research projects: Sponsored research charges model usage to an approved grant or project chartstring.
  • Other UC campuses: Intercampus recharge agreements are arranged via tritonai@ucsd.edu.
  • Health System patient care: Clinical operations must use approved Medical Center AI platforms on Pulse.

Finish setup and try TritonAI Harness

Let the installer finish setup, then launch TritonAI Harness and try a prompt that contains no sensitive information.

  1. 1
    Enter your key and install

    Paste your key into TritonAI Installer and choose Check access & install. Keep it open until setup finishes.

  2. 2
    Open TritonAI Harness

    Launch the app and choose a model available to your key. The managed default is GLM 5.3 Flash. TritonAI Harness starts with Auto when there’s no saved override or remembered choice. Before sending a request, select Supervised in the message composer so file changes and write tools pause for approval.

  3. 3
    Try a first prompt

    Ask “Help me outline a meeting agenda.” Confirm you receive a response, then review it.

  4. 4
    Learn the workspace

    Start with Essentials in the standalone course, then create and check a feedback summary using the fictional practice kit.

    Start TritonAI Harness training

Need setup help?

If your key is rejected or models do not appear, remove extra spaces around your key. Keep the key private; support will never ask for your secret key.

Email tritonai@ucsd.edu with your operating system and error details. Never include the secret key in email.

Updates and optional computer use

For an existing TritonAI Harness installation, open Settings > About > Check for Updates. Version 0.4.1 is the current stable release. Stable and nightly installations use separate update tracks. The guided installer above is the first-time setup and runtime repair path.

Computer use starts off. To enable it, open Settings > General and follow the readiness instructions. On macOS, grant Accessibility and Screen Recording permissions, then restart the TritonAI Harness when prompted. In this stable release, computer use requires the Codex provider on a local desktop backend. Screenshots may be included in model requests; use approved data and review the TritonAI Harness privacy statement.

Other clients and direct API use

Using another client or your own code? Request a Gateway key, then configure your chosen client. You do not need TritonAI Installer for this path.

CLI Coding Agents

Claude Code, Codex, Hermes, and other clients

Claude Code and Codex are supported alternatives. Follow the client setup guides for the endpoint and settings your client needs.

  • Base URL: https://tritonai-api.ucsd.edu/v1
  • Default Model: api-glm-5.3
  • Keep your API key in a secure local credential store or environment variable. Never commit it to code.
API & Pipelines

Direct Code Integration

Use the OpenAI Python SDK.

from openai import OpenAI
client = OpenAI(
  base_url="https://tritonai-api.ucsd.edu/v1",
  api_key="YOUR_KEY"
)
resp = client.chat.completions.create(
  model="api-glm-5.3",
  messages=[{"role":"user","content":"Hi"}]
)
Follow the API quickstart for Python and cURL examples. See TritonAI model catalog for available routes.