Quickstart

This guide takes you from your first sign-in to an evaluated trace. It follows the Custom system integration path, which connects your own agent. If you would rather see AgentHub work on a pre-built example first, pick Quickstart with pre-built templates in step 2 and skip to Cost Calculator afterwards.

Before you start

Your email must be approved for platform access. Sign-in is allowlisted. If your address, or your whole email domain, has not been approved, sign-in stops with “Your access is pending approval by an administrator.” Ask one of your organization’s super admins to approve you. See Platform access.
A provider API key from Anthropic, OpenAI, or Google, if you want to bring your own. You can start on the Platform Demo Key instead.
A way to send an HTTP request, such as curl, to post your first trace.

Steps

Go to https://pillarforge.corepillar.ai. Enter your work email and select Send me a code. Enter the 6-digit code from the email and select Verify & sign in.The first time you sign in, AgentHub creates your account and, because you belong to no organization yet, sends you to onboarding.If your organization uses Okta, a Sign in with SSO button appears once you type your email. See SSO with Okta.

Create your organization and project

On How would you like to get started?, choose Custom system integration (Connect my system). The path has five steps:

Connect system. Pick REST ingest API under “How will your system connect?” and pick Sandbox or Production.
LLM keys. Choose Use the Platform Demo Key or Bring my own provider key. If you bring your own, pick the provider and paste the key. It is held in the browser tab until your organization exists.
Create org. Enter an Organization name and select Create organization. You become its Owner.
Create project. Enter a Project name and select Create project.
Upload instructions. Optionally drop in your agent’s Markdown instructions, then select Save and finish or Finish without instructions.

You land in Model Manager. Every field is described in Onboarding.

Add a provider key and a model

In Model Manager:

If you skipped your own key, the API Key Vault already holds the Platform Demo Key, a platform-funded allowance of $50.00 on Anthropic models. To add your own key, pick a Provider, paste the API key, optionally add a Label (optional), and select Save key. AgentHub validates the key with the provider before it stores it encrypted, and never shows it again.
Under Add model, choose the provider, model, and the key to bill it to, then select Add model.

If you saved your own key during onboarding, AgentHub already added that provider’s default model to the project: GPT-4o Mini for OpenAI, Claude Haiku 4.5 for Anthropic, or Gemini 3.5 Flash-Lite for Google.A model is required to run LLM-judge evaluators later. See Model Manager.

Create a tracing key

Open Settings, then Tracing keys. Choose your project and an environment (Live or Test), then select + New live key (or + New test key). Copy the key from the New tracing key window. It is shown once.

The Tracing keys tab is visible only to your platform’s super admins. If you are an org Owner or Admin and don’t see it, ask a super admin to create the key for you. See Tracing keys.

Send your first trace

Post a trace with one span. Replace the key with yours.

curl -X POST https://pillarforge.corepillar.ai/api/v1/traces/ingest \
  -H "Authorization: Bearer tk_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "trace_id": "quickstart-001",
    "root_name": "agent_run",
    "environment": "development",
    "spans": [
      {
        "span_id": "span-1",
        "name": "chat.completions.create",
        "kind": "llm",
        "model": "gpt-4o-mini",
        "status": "OK",
        "tokens_in": 120,
        "tokens_out": 45,
        "cost_usd": 0.00012,
        "latency_ms": 890,
        "input": { "prompt": "What is our refund window?" },
        "output": { "text": "Refunds are accepted within 30 days." },
        "started_at": "2026-10-19T10:00:00.000Z",
        "ended_at": "2026-10-19T10:00:00.890Z"
      }
    ]
  }'

A successful request returns 202. The request format, limits, and retry behavior are in the tracing quickstart.

Find the trace

Open Traces in the sidebar. Your trace appears with its name, status, span count, tokens, and cost. Select it to open the detail view. See Traces.

Evaluate it

On the trace detail page, the Evaluate trace panel offers Evaluate this trace. It opens Evaluation on the Your Evaluations tab with the trace and your first collection already filled in. Run it to get a score, a pass or fail, and a reason for each evaluator.If the panel says “Create an evaluator collection first”, select Go to Evaluation, build a collection from the built-in evaluators, and come back. See Evaluation runs.Running evaluations needs the AI Builder role or higher. Viewers see “You need Editor access to run evaluations on this trace.”

Where to go next

Invite your team

Add admins, AI Builders, and ModelHub users.

Trace format

Every field the ingest endpoint accepts.

Evaluators

Built-in and custom evaluators.

Cost Calculator

Compare models on the same scenarios.

Get started

Platform

Evaluate and observe

Administration

Before you start

Steps

Where to go next

Invite your team

Trace format

Evaluators

Cost Calculator

​Before you start

​Steps

​Where to go next

Invite your team

Trace format

Evaluators

Cost Calculator

Before you start

Steps

Where to go next