Confirm an owner
Assign a workspace or project owner who can approve models, review usage, and respond when a request is blocked.
Integrations
Connect the tools your team already uses.
Point compatible clients and API workflows at New API once. Keep familiar tools while your organization controls model access, usage, and spend.
Before you connect
Prepare the access path first.
A reliable integration starts with a clear owner, an approved model, and a key that can be revoked without changing application code.
Approve the model
Check the model catalog and confirm that the capability, context, and provider route fit the intended workflow.
Set a quota
Give the key or team an explicit usage boundary before connecting a shared editor, agent, or service.
Compatible clients
One managed route for everyday AI work.
Use New API from the editor, terminal, or compatible API client that fits the workflow.
VS Code
Editor-based AI workflows
Codex
Agent and terminal workflows
Claude Code
Anthropic-compatible workflows
Trae
AI-first editor workflows
Zcode
Coding assistant workflow
Compatible clients
OpenAI-compatible configuration
Configure with care
Keep credentials and request behavior predictable.
The integration should be easy for a developer to test and easy for an administrator to review.
Use a server-side key
Store the New API API key in an environment variable or secret manager. Never place it in browser code, a repository, or request logs.
Match the request format
Choose the endpoint that matches the model family, then keep the model identifier and required fields in the request body.
Test a small request
Start with a representative, low-volume request. Confirm the response, usage record, quota behavior, and error handling before rollout.
Agent orchestration
Let agents call tools through one governed route.
An agent can plan a task, retrieve context, and call an approved model while New API keeps the credentials, model access, and request trail under the same operational controls.
Task request
Agent
Retrieve context
Model call
Reviewed result
A clear integration path
From configuration to governed output.
- 01
Choose an approved model
Select a model and capability your organization has made available.
- 02
Set the compatible endpoint
Configure the client with a New API endpoint and managed API key.
- 03
Review usage in context
Use request records, quotas, and usage data before expanding access.
Integration FAQ
Practical answers before production.
Use these answers to plan a first connection. The API Reference remains the source of truth for endpoint fields and provider-specific behavior.
Where should we start?
Start with one owned workflow, one approved model, and one test key. Connect it in a non-production environment, verify the response and usage record, then expand deliberately.
How do we choose the base URL and endpoint?
Use the New API base URL provided for your workspace and select the compatible endpoint for the model family. The API Reference lists the current paths, required headers, and request fields.
Where should the API key live?
Keep it on the server or in the client tool secret store, supplied through an environment variable where possible. Do not expose it in frontend bundles, source control, screenshots, or logs.
Are OpenAI, Anthropic, and Gemini formats interchangeable?
They are not interchangeable for every model. New API supports compatible formats where applicable, including Chat Completions, Responses, Anthropic Messages, and Gemini GenerateContent. Check the model and endpoint documentation before switching formats.
How do we know which models are available?
Use the model catalog and your workspace permissions together. A model can appear in the catalog but still require an administrator to authorize it for your team or key.
What should we do with 401, 403, 404, or 429 errors?
Check the key and header for 401, permissions or model access for 403, the path and model compatibility for 404, and quota or rate limits for 429. Preserve the request ID and review the API Reference or support channel for persistent failures.
How should we handle limits in production?
Set quotas before launch, use bounded retries with backoff for transient responses, and monitor usage and error rates. Do not retry authentication or permission failures blindly.
When should we read the API Reference?
Use it whenever you need exact endpoint paths, headers, request schemas, streaming behavior, media fields, or provider-specific constraints. This page explains the integration workflow; the reference defines the contract.