All major providers, one integration

Switch AI providers by changing one parameter, since Inference sends each request to the provider of the model that you specify in it. Inference can also send one prompt to several models at once so that you can compare their answers.

Get early accessGet in Touch

One API for all major AI providers

Inference returns the response from every provider in the same format, so the code that handles a response stays the same when you change models. Through the same API, you can compare several models on one prompt, sample one model several times, or generate images and video into Storage.

Ways to call and combine models

Multi-provider support

Inference connects your application to all major AI providers through one API. The documentation lists the models that each provider offers through Conjoin in the current release.

Parallel execution

Conjoin calls every model on a multi-model request in parallel and returns their responses together, ready for you to compare or merge.

Best-of-N sampling

Ask one model for up to four samples of the same prompt. Conjoin can then pick the sample that best meets a criterion that you set, such as coherence, or it can merge all the samples into one answer.

Multi-modal generation

Inference saves each image or video that a model generates to a Storage path of your choosing. For analysis, pass the Storage path of an image, an audio file, or a video in place of the file itself.

Response synthesis

With synthesis on, a judge model evaluates the response from every model on the request, then either keeps the best one or combines them into one answer.

Usage tracking

Every Inference response includes its request ID, its token usage, its latency in milliseconds, and the charge for the request in US dollars. Conjoin also totals that usage by project, team, or user.

All major
AI providers

Through a single Inference API

4x
Multi-sample

Per-model sample limit

< 100ms
Routing overhead

Latency that Conjoin adds to a multi-model request

Build faster with solutions designed for real-world needs

Reach all major providers through one integration

Pass a different value in the model parameter to send the request to another provider. Your code then extracts the text from choices[0].message.content whichever model answers.

Impact

In the sample, one function requests the Plover Robotics release note from either model, with the model argument as the only difference between the two calls.

Prompt for a coding agent: Write the release note with either model, and let us switch models with one argument.

Conjoin IDE

Select a file to view its contents

Cross-check an answer across models or across samples

Compare how several models answer one prompt, or how one model answers the same prompt several times. When you synthesize the responses, the judge model marks the points where they agree and flags the points where they disagree so that a person can review the difference.

Impact

In the sample, Conjoin selects the most coherent of four drafts that one model writes for a Kestrel Underwriting rationale. The code then saves the request ID, the token count, and the cost so that a reviewer can trace the decision.

Prompt for a coding agent: Have one model write four drafts of the rationale, and keep the most coherent one.

Conjoin IDE

Select a file to view its contents

Generate images into a Storage path

Inference writes each generated image to the Storage container that you name, then returns the file's Storage path and a CDN URL.

Impact

Your application can display a new image from its CDN URL as soon as Inference responds, because Conjoin stores the file before it sends the response.

Point a vision model at an image in Storage

Put an image's Storage path in the prompt for a vision model. Conjoin checks that the requesting user can open the file before it fetches the image from Storage and encodes it for the model.

Impact

Your vision feature can analyze the images that are already in Storage with a single API call.

Build your next app on Conjoin

Conjoin will open with a free Build tier, while the pricing page already lists the launch prices for Pro and Business. If you tell us what you're building, we'll talk you through early access.

See pricingGet early access