PipeShift

PipeShift provides an OpenAI-compatible chat completions API for open-source and custom models. Use Custom Endpoints in the AI Gateway to register PipeShift once, keep the PipeShift API key on the gateway, and let your applications call models through a single TrueFoundry API key with access control and tracing.

Prerequisites

Create a PipeShift account

Generate a PipeShift API key

In the PipeShift dashboard, create an API key for your workspace.Store the key securely. You will paste it into the gateway’s Custom Headers configuration — not in client application code.

Get your TrueFoundry API key and gateway URL

You need a TrueFoundry API key (TFY_API_KEY) and your gateway base URL to call models through the gateway. See Gateway base URL and Authentication.

Adding models

Add PipeShift to the AI Gateway using Custom Endpoints.

Create a Custom Endpoint provider account

In the TrueFoundry dashboard, go to AI Gateway → Models → Custom Endpoints and click Add Custom Endpoint.In Configure Account, set:

Name (Account Name): any unique label for this provider (for example, pipeshift) — this is your provider account name and appears as the first path segment in every request URL ({providerAccountName})
Endpoint Type: None
Header Auth: keep disabled

Add a PipeShift endpoint

On the Endpoints step, add an integration and configure:

Display Name: any unique label for this endpoint (for example, model) — this is your custom endpoint display name and appears as the second path segment in every request URL ({endpointName})
Base URL: https://api.pipeshift.com/api

Enable Custom Headers and add:

Content-Type: application/json
Authorization: Bearer <PIPSHIFT_API_KEY>

Keep Header Auth and TLS Settings disabled unless your deployment requires them.

Custom Endpoint form for PipeShift with Base URL and custom headers — Endpoint settings — Base URL and upstream headers

The Authorization header here is sent from the gateway to PipeShift. Your application should only send the TrueFoundry API key to the gateway.

You can click + Add Custom Endpoint to register multiple endpoints under the same account (for example, model, vision, or fast-inference). Each gets its own Display Name in the URL, but they can all point to the same PipeShift Base URL and headers. Which PipeShift model runs is chosen in the request body model field — not by the display name.

Set access control and save

On the Access Control step, choose who can manage and use this provider account, then Save.

Manager: users/teams who can edit or delete the custom endpoint
User: users/teams who can call the endpoint (for example, everyone)

Inference

Once saved, call PipeShift through the gateway’s proxy-api path. URL shape and path rules are documented under Custom Endpoints.

How the request URL is built

The gateway URL has two values you configure in the dashboard — they are not PipeShift model IDs:

URL segment	Dashboard field	Example in this guide
`{providerAccountName}`	Account Name (Configure Account → Name)	`pipeshift`
`{endpointName}`	Custom endpoint Display Name (Endpoints step)	`model`

Full URL pattern:

{GATEWAY_BASE_URL}/proxy-api/{providerAccountName}/{endpointName}/v0/chat/completions

With the example configuration above:

{GATEWAY_BASE_URL}/proxy-api/pipeshift/model/v0/chat/completions

Segment	Example value	Meaning
`{GATEWAY_BASE_URL}`	`https://internal.devtest.truefoundry.tech/api/llm`	Your AI Gateway base URL
`pipeshift`	Account Name you set in Configure Account	`{providerAccountName}`
`model`	Display Name you set for the custom endpoint	`{endpointName}`
`v0/chat/completions`	Upstream path	Appended to the integration Base URL

The gateway forwards the request to:

https://api.pipeshift.com/api/v0/chat/completions

Choosing a PipeShift model

PipeShift hosts many models available in your workspace. You do not need a separate custom endpoint per model — set the model field in the JSON request body to whichever PipeShift model ID you want to run (for example, zai-org/GLM-5). If you add multiple custom endpoints under the same account (different Display Name values), use the matching {endpointName} in the URL; each endpoint can still call any PipeShift model via the request body.

Replace pipeshift and model in the URL with your own Account Name and Display Name if you used different values during setup.

Supported APIs

API	Endpoint	Tracing	Cost Tracking
Chat Completions	`/proxy-api/{providerAccountName}/{endpointName}/v0/chat/completions`	✅

Chat Completions

Before you start: Replace {GATEWAY_BASE_URL} with your gateway base URL (how to find it) and set TFY_API_KEY to your TrueFoundry API key.

Request headers (client → gateway)

Header	Value
`Authorization`	`Bearer <TFY_API_KEY>`
`Content-Type`	`application/json`
`Accept-Encoding`	`identity`

Python (`requests`)

import os

import requests

GATEWAY_BASE_URL = "{GATEWAY_BASE_URL}"  # e.g. https://internal.devtest.truefoundry.tech/api/llm
TFY_API_KEY = os.environ.get("TFY_API_KEY", "your-tfy-api-key")

# providerAccountName = Account Name from Configure Account (e.g. pipeshift)
# endpointName = Custom endpoint Display Name (e.g. model)
# model in BODY = PipeShift model ID (e.g. zai-org/GLM-5)
PROVIDER_ACCOUNT_NAME = "pipeshift"
ENDPOINT_NAME = "model"
URL = f"{GATEWAY_BASE_URL}/proxy-api/{PROVIDER_ACCOUNT_NAME}/{ENDPOINT_NAME}/v0/chat/completions"

HEADERS = {
    "Authorization": f"Bearer {TFY_API_KEY}",
    "Content-Type": "application/json",
    "Accept-Encoding": "identity",
}

BODY = {
    "model": "zai-org/GLM-5",  # any PipeShift model ID
    "messages": [
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Tell me about Truefoundry AI Gateway"},
    ],
    "temperature": 0.7,
    "stream": False,
}

response = requests.post(URL, headers=HEADERS, json=BODY, timeout=120)
response.raise_for_status()

print(response.json())

cURL

curl -X POST "{GATEWAY_BASE_URL}/proxy-api/pipeshift/model/v0/chat/completions" \
  -H "Authorization: Bearer $TFY_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Accept-Encoding: identity" \
  -d '{
    "model": "zai-org/GLM-5",
    "messages": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user", "content": "Tell me about Truefoundry AI Gateway"}
    ],
    "temperature": 0.7,
    "stream": false
  }'

OpenAI Python SDK

Because PipeShift is OpenAI-compatible, you can point the OpenAI SDK at the gateway proxy path instead of PipeShift directly:

import os

from openai import OpenAI

GATEWAY_BASE_URL = "{GATEWAY_BASE_URL}"  # e.g. https://internal.devtest.truefoundry.tech/api/llm
TFY_API_KEY = os.environ.get("TFY_API_KEY", "your-tfy-api-key")

PROVIDER_ACCOUNT_NAME = "pipeshift"  # Account Name
ENDPOINT_NAME = "model"             # Custom endpoint Display Name

client = OpenAI(
    api_key=TFY_API_KEY,
    base_url=f"{GATEWAY_BASE_URL}/proxy-api/{PROVIDER_ACCOUNT_NAME}/{ENDPOINT_NAME}/v0",
)

response = client.chat.completions.create(
    model="zai-org/GLM-5",  # any PipeShift model ID
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Tell me about Truefoundry AI Gateway"},
    ],
    temperature=0.7,
    stream=False,
)

print(response.choices[0].message.content)

Support scope: Custom Endpoints proxy requests transparently to PipeShift. For gateway limitations on Custom Endpoints (HTTPS, streaming, and so on), see Custom Endpoints.

Get Started

LLM Gateway

MCP Registry and Gateway

Skills Registry

Prompt Registry

Guardrails and Security

Observability

Deployment

Admin Guide

Chat

Messages

Embeddings

Rerank

Responses

Image

Audio

Batch

Files

Fine-tuning

Moderations

Models

Prerequisites

Adding models

Inference

How the request URL is built

Choosing a PipeShift model

Supported APIs

Request headers (client → gateway)

Python (`requests`)

cURL

OpenAI Python SDK

​Prerequisites

​Adding models

​Inference

​How the request URL is built

​Choosing a PipeShift model

​Supported APIs

​Request headers (client → gateway)

​Python (requests)

​cURL

​OpenAI Python SDK

Prerequisites

Adding models

Inference

How the request URL is built

Choosing a PipeShift model

Supported APIs

Request headers (client → gateway)

Python (`requests`)

cURL

OpenAI Python SDK