> ## Documentation Index
> Fetch the complete documentation index at: https://docs.corepillar.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Overview

> The project home page in AgentHub: setup options, the OpenTelemetry integration guide, project stat tiles, and the most recent runs.

Overview is the first item in the AgentHub sidebar and the page you land on when you open a project. It answers three questions for the project you have selected: how do I get started, how do I send traces, and what has this project done so far.

The page header reads **Overview**, followed by "Get *project name* set up, then track models, evaluations, and spend as they happen."

The page has four sections, top to bottom:

1. **Set up your first evaluation** (two setup options)
2. **Send your first trace** (the integration guide)
3. Stat tiles: **Models Configured**, **Evaluations Run**, **Total Cost Analyzed**
4. **Recent Runs**

The first two sections render immediately. The stat tiles and the runs table load in the background and show a skeleton placeholder until their data arrives.

## Set up your first evaluation

This card offers two ways to start, separated by an **OR** badge.

### Option 1: Connect Your Model

For when you want to evaluate your own agent with your own provider key.

The card lists the steps:

1. Open **Model Manager** from the sidebar
2. Add your provider + API key
3. Select a project and environment

The **Go to Model Manager** button opens [Model Manager](/agenthub/model-manager). If you already have a model configured, the link **Jump straight to Cost Calculator.** skips ahead to the [Cost Calculator](/agenthub/cost-calculator).

### Option 2: Try a Project Template

"Fully configured agents with realistic mock data. One click to try." Four featured templates are shown as cards. Each card shows the template name, a one-line description, and the connectors the agent uses.

| Card | What the agent does | Connectors |
| - | - | - |
| Conference Lead Deduplication | A field-marketing agent that dedupes badge-scanned event leads. One merge fails mid-run on a one-digit phone mismatch, so the trace shows the failure, the recovery, and the work that carried on. | Swapcard Events, HubSpot CRM, Clearbit Enrichment |
| Order to Cash Automation | A revenue operations agent that reconciles payments, evaluates credit risk, and stages ledger adjustments. | NetSuite ERP, Stripe Billing |
| Automated Hiring Process | A recruiting operations agent that screens candidates against compensation bands and stages offer letters. | Greenhouse ATS, Workday, Paradox Olivia, DocuSign |
| Customer Support & Refund Resolution | A support agent that resolves refund requests within policy and escalates edge cases to a supervisor. | Zendesk, Stripe |

Clicking a card does not change your current project. It creates a **new sibling project** in the same organization, named after the template, with the template's system prompt, tone, skills, memory and tools saved as a draft agent configuration. You are then taken to that new project's Cost Calculator with the form already filled in. A toast confirms `"<template name>" project created`. While the request runs, the card reads **Creating…** and the other cards are blocked.

<Note>
  Only organization admins can create projects. If your role is not Owner or Admin (and does not hold the `manage_org_settings` permission), the cards are dimmed, the card shows "Only an org admin can create a project from a template.", and clicking one shows "Ask an org admin to start a project from a template." See [Roles and permissions](/agenthub/settings/roles-and-permissions).
</Note>

The full template library (ten templates, including the Coding Agent Template) is available from the **Start from…** menu inside the Cost Calculator, which fills the current project's form instead of creating a new project. See [Cost Calculator templates](/agenthub/cost-calculator/templates).

## Send your first trace

This section is the integration guide for teams that already run OpenTelemetry. AgentHub receives OTLP over HTTP, so you point your existing exporter at the endpoint and your spans appear under [Traces](/agenthub/traces). There is no proprietary SDK to install and no change to your call sites.

The guide lists three steps:

1. Create a tracing key under **Settings → Tracing Keys** (links to [Tracing keys](/agenthub/settings/tracing-keys))
2. Point your exporter at the endpoint and send the key as a bearer token
3. Run your agent. Spans arrive in Traces.

### Connection details

| Row | Value shown | Note on the card |
| - | - | - |
| OTLP endpoint | `https://pillarforge.corepillar.ai/api` | "Exporters append /v1/traces themselves." |
| Auth header | `Authorization: Bearer tk_live_...` | "One key per project, created in Settings." |

Each row has a **Copy** button. After a successful copy it reads **Copied**. If the browser blocks clipboard access it reads **Select to copy** so you can select the text by hand.

The endpoint is the base URL. Your exporter adds `/v1/traces` to it, so traces are posted to `https://pillarforge.corepillar.ai/api/v1/traces`.

<Info>
  The snippets always contain the placeholder `tk_live_...`, never a real key. Raw tracing keys are shown once, when you create them, and only a hash and a short hint are stored. Paste your own key over the placeholder.
</Info>

### Snippet tabs

The tabs give one runnable setup per language. Each tab has a short description and one or more code blocks, each with its own **Copy** button.

| Tab | Description on the tab | Code blocks |
| - | - | - |
| Python (marked **Recommended**) | OpenLLMetry auto-instruments the OpenAI, Anthropic, Bedrock, LangChain and LlamaIndex clients you already call. Nothing at your call sites changes. | `install` (`pip install traceloop-sdk`), `main.py` |
| Python · OTel | For teams already running OpenTelemetry who would rather not add another dependency. | `install`, `tracing.py` |
| TypeScript | Standard OpenTelemetry Node SDK, loaded before your app so it can patch modules as they load. | `install`, `instrumentation.ts` |
| Env vars | Zero code, any language. Every OpenTelemetry SDK reads these, so an already-instrumented service only needs the variables set. | `.env`, `otel-collector-config.yaml` |
| Go | The stock OTLP/HTTP trace exporter. | `install`, `tracing.go` |
| Java | The OpenTelemetry Java agent instruments a JVM app without recompiling it. | `download the agent`, `run` |

The Env vars tab sets `OTEL_EXPORTER_OTLP_PROTOCOL=http/protobuf`. Several SDKs default to gRPC, which this endpoint does not accept.

<Tip>
  The OpenLLMetry and OpenTelemetry SDKs also start metric and log exporters aimed at the same base URL. AgentHub accepts those requests (authenticated with the same tracing key) so your console does not fill with 404 errors, but it does not store metrics or logs. Only traces are kept.
</Tip>

### Tracing key status

Below the tabs, a status line tells you whether the project already has a key:

| State | What you see | Button |
| - | - | - |
| No active key | "No tracing key for this project yet — create one, then paste it over `tk_live_...`." | **Create a tracing key** |
| Active key | "This project's live key:" (or "test key:") followed by the masked key, and "The full value is only shown once, at creation." | **Manage tracing keys** |
| Could not load | "Couldn't load this project's tracing keys just now." | **Create a tracing key** |

Only the newest non-revoked key is described. Both buttons open [Settings → Tracing Keys](/agenthub/settings/tracing-keys). For the full ingestion contract, see the [tracing quickstart](/agenthub/developers/tracing-quickstart).

## Stat tiles

Three tiles summarize the selected project. They are always project-scoped, never org-wide.

| Tile | What it counts |
| - | - |
| **Models Configured** | Model configurations registered to this project in [Model Manager](/agenthub/model-manager). |
| **Evaluations Run** | Trace evaluation runs (running, completed or failed) plus Cost Calculator runs that actually executed (running, complete or failed). Cost Calculator drafts and runs that only generated scenarios are not counted. |
| **Total Cost Analyzed** | A monthly projection, not measured spend. See below. |

### How Total Cost Analyzed is calculated

1. AgentHub sums the recorded cost of every trace in the project, live traffic and Cost Calculator simulations alike, and counts those traces. A simulated scenario whose trace is missing is counted from its own result row instead, so no dollar is counted twice.
2. It divides cost by count to get a **cost per request**.
3. It multiplies that rate by a monthly request volume.

The default volume is the Cost Calculator's default scale: **5,000 monthly users × 3 requests per user = 15,000 requests a month**. If you have changed the scale for this project in the Cost Calculator's **Edit scale** dialog, the tile uses your project's figure instead. That setting is remembered in your browser, so a colleague on another machine sees the default until they set it too.

The caption under the tile always says which basis it used, for example:

* "Projected at 5K users/mo — the platform default; this project has not set its own · from 42 recorded agent runs"
* "Projected at ... — this project's configured volume · from ..."
* "Nothing recorded yet — run an evaluation to project spend" (when the project has no recorded cost yet; the tile shows \$0.00)

Figures under \$1 show cents. Larger figures are rounded to whole dollars.

## Recent Runs

A table of the six most recent runs in the project, merged from two sources and sorted newest first:

* **Cost Calculator** runs, including drafts and runs that only generated scenarios
* **Trace Evaluation** runs (a collection applied to a single trace)

| Column | Content |
| - | - |
| **Run** | The first 8 characters of the run ID (hover for the full ID), with the run type underneath: "Cost Calculator" or "Trace Evaluation". The ID is a link. |
| **Status** | Badge: Completed, Failed, Running, Draft or Generated |
| **Model** | For Cost Calculator runs, the model under test (catalog name when known). For trace evaluations, the judge model. "—" when there is none. |
| **Cost** | For Cost Calculator runs, measured cost if the run executed, otherwise the pre-run estimate. For trace evaluations, the recorded cost of the trace that was graded (not the cost of the judge calls). "—" when no figure exists. |
| **Created** | Local date and time |

Clicking a Cost Calculator run opens it in the Cost Calculator (`/cost-calculator?run=<id>`), landing on its results when it has executed. Clicking a trace evaluation opens the trace it graded in [Traces](/agenthub/traces).

**Empty state:** "No runs yet — Start your first evaluation above — connect a model or try a project template."

## Who can see this page

Any AgentHub user who is a member of the organization can open Overview. Users with the Viewer role are ModelHub-only and see "Your account has ModelHub access only" instead of AgentHub pages. See [Roles and permissions](/agenthub/settings/roles-and-permissions).

## Related

<CardGroup cols={2}>
  <Card title="Model Manager" icon="key-round" href="/agenthub/model-manager">
    Store provider keys and register models for this project.
  </Card>

  <Card title="Cost Calculator" icon="calculator" href="/agenthub/cost-calculator">
    Generate scenarios and measure real cost and quality per model.
  </Card>

  <Card title="Traces" icon="activity" href="/agenthub/traces">
    Where the spans you send show up.
  </Card>

  <Card title="Tracing keys" icon="key" href="/agenthub/settings/tracing-keys">
    Create the key your exporter authenticates with.
  </Card>
</CardGroup>


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.