# Install and configure

Install the SDK and configure a managed or self-hosted gateway.

Source: https://tokentriage.com/docs/sdks/install

---
import NextSteps from '../../../components/NextSteps.astro';

The configured clients in this documentation are **unreleased**.
Use this checkout to try them. The earlier registry release candidate does
not contain these APIs.

## Python

From the repository root:

```console
python -m pip install "./sdk/python[openai]"
```

The OpenAI extra installs the provider integration. The instrumentation core
has no runtime dependencies. Python 3.10 or newer is required.

## TypeScript

The following commands use Node 24 to run TypeScript directly. Start at the
repository root, build the package, and prepare a small backend application:

```console
cd sdk/ts
npm ci
npm run build
npm pack
cd ../..
mkdir -p test-results/sdk-example-app
cd test-results/sdk-example-app
npm init -y
npm pkg set type=module
npm install ../../sdk/ts/tokentriage-1.0.0-rc.2.tgz openai@7.10.0
cp ../../sdk/examples/first_request.ts first_request.ts
cd ../..
```

PowerShell:

```powershell
Set-Location sdk/ts
npm.cmd ci
npm.cmd run build
npm.cmd pack
Set-Location ../..
New-Item -ItemType Directory -Force test-results/sdk-example-app | Out-Null
Set-Location test-results/sdk-example-app
npm.cmd init -y
npm.cmd pkg set type=module
npm.cmd install ../../sdk/ts/tokentriage-1.0.0-rc.2.tgz openai@7.10.0
Copy-Item ../../sdk/examples/first_request.ts first_request.ts
Set-Location ../..
```

Both sequences return to the repository root. The installed tarball contains the
local SDK; the application also has the official OpenAI package. Existing Node 22
applications can use their TypeScript compiler or runner. The instrumentation core
supports Node 18 or newer.

## Set your managed connection

Your TokenTriage account key belongs in `X-Tokentriage-Key`.
Your provider key stays in its provider credential slot.
The configured client supplies both at request time.

Replace these values with your assigned URL, keys and enabled model:

```console
export TOKENTRIAGE_BASE_URL='your-assigned-gateway-url'
export TOKENTRIAGE_API_KEY='your-account-key'
export OPENAI_API_KEY='your-provider-key'
export TOKENTRIAGE_MODEL='your-enabled-model'
```

```powershell
$env:TOKENTRIAGE_BASE_URL = 'your-assigned-gateway-url'
$env:TOKENTRIAGE_API_KEY = 'your-account-key'
$env:OPENAI_API_KEY = 'your-provider-key'
$env:TOKENTRIAGE_MODEL = 'your-enabled-model'
```

Keep credentials in your backend.
From the repository root, run the example for your language:

```console
python sdk/examples/first_request.py
```

```console
node test-results/sdk-example-app/first_request.ts
```

Each prints the answer and request ID.
See [the first-request code and failure guidance](/docs/sdks/first-request).

## Self-hosted configuration

Use your deployment's URL, including any path prefix, and select
`mode="self_hosted"` in Python or `mode: "self_hosted"` in TypeScript.
Unset `TOKENTRIAGE_API_KEY` if your gateway does not require it.
Supply the upstream credential required by your own deployment.

## Verify before you spend

Onboarding may supply a `health_url`. A GET to that exact URL checks gateway
reachability without inference. It does not verify provider credentials or models.

The legacy `check()` helper probes an admin endpoint. It is useful for
self-hosted administration and is not required for a managed first request.

<NextSteps items={[
  { title: 'First routed request', href: '/docs/sdks/first-request' },
  { title: 'Provider clients and streaming', href: '/docs/sdks/adapters' }
]} />