Inference - Early Access

Warning

This guide describes the process of accessing the early access Inference service. Please note that this service is early access, under active development and not designed for production workloads.

Access must be requested

At this time, access to this service is on an invite-only basis. Being a member of an Isambard-AI project does not give you access to the Inference service. If you have not been invited, the steps in this guide will not work for you.

Introduction¶

The BriCS Inference service is available for early access users to test and provide feedback. For this we provide an OpenAI-compatible API following the /v1/chat/completions/ standard. We currently host a set-menu of models, that is expandable upon request.

The API endpoint, authentication, and models are described below.

API Endpoint¶

The endpoint is available here:

https://apps.isambard.ac.uk/inference/v1

Authentication - Getting an API token¶

All calls to the API must provide an HTTP-Bearer credential.

To authenticate and obtain an API token you must use the chepstow CLI client (similar to clifton).

Install Chepstow¶

To install chepstow, follow the tabs to your respective operating system:

macOSLinux (including WSL)

Download the binary of Chepstow using curl:

Apple Silicon (Arm)Intel

curl --location https://github.com/isambard-sc/chepstow/releases/latest/download/chepstow-macos-aarch64 --output chepstow
chmod u+x chepstow
# Place in local/bin or your preferred location
mv chepstow ~/.local/bin/

curl --location https://github.com/isambard-sc/chepstow/releases/latest/download/chepstow-macos-x86_64 --output chepstow
chmod u+x chepstow
# Place in local/bin or your preferred location
mv chepstow ~/.local/bin/

If you want it accessible from any directory on your computer, you can place the binary in ~/.local/bin/ as above

If you don't, you will need to specify the path to the executable when running it. So instead of chepstow auth as described below you would run e.g. ./chepstow auth or ~/chepstow auth.

Allowing Chepstow to run on macOS

If you download the Chepstow through your web browser rather than with curl, you may be presented with a macOS warning dialog when trying to run the executable, e.g.

"chepstow" can't be opened because Apple cannot check it for malicious software.

If this occurs you will have to go into "System Settings > Privacy & Security" and allow use of the chepstow executable (see Open a Mac app from an unidentified developer from the macOS documentation). Note that you will need to have admin privileges to change the settings in "Privacy & Security".

Download the binary of Chepstow using curl:

x86_64AArch64

curl --location https://github.com/isambard-sc/chepstow/releases/latest/download/chepstow-linux-musl-x86_64 --output chepstow
chmod u+x chepstow
# Place in local/bin or your preferred location
mv chepstow ~/.local/bin/

curl --location https://github.com/isambard-sc/chepstow/releases/latest/download/chepstow-linux-musl-aarch64 --output chepstow
chmod u+x chepstow
# Place in local/bin or your preferred location
mv chepstow ~/.local/bin/

If you want it accessible from any directory on your computer, you can place the binary in ~/.local/bin/ as above

If you don't, you will need to specify the path to the executable when running it. So instead of chepstow auth as described below you would run e.g. ./chepstow auth or ~/chepstow auth.

Authenticate¶

Use the following steps after installing chepstow:

chepstow auth

Follow the link in your browser to login, choose your group if you are in multiple projects. You can then get your token with this command:

chepstow token

You can then use this API Bearer token with your normal harnesses and tools.

API tokens are limited to 24 hours

API token lifetimes are limited to 24 hours. You must re-run chepstow auth after expiry.

Models list¶

Chepstow can also help you list the models available:

$ chepstow models

For example this will print:

qwen3-8b
deepseek-v4.1-flash

Usage¶

You can use the API just like any other OpenAI-compatible endpoint. The information you need is the base url, a Bearer API Token from chepstow, and a model:

Provider	OpenAI Compatible
Base URL	https://apps.isambard.ac.uk/inference/v1
API Key	eyJhb... or $(chepstow token)
Model	deepseek-v4.1-flash

Below we provide examples of using this with a coding harness.

OpenCode¶

Here we provide an example config to use the inference service with the OpenCode harness.

Place this content in ~/.config/opencode/opencode.jsonc:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "isambard": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Isambard",
      "options": {
        "baseURL": "https://apps.isambard.ac.uk/inference/v1"
      },
      "models": {
        "deepseek-v4.1-flash": {
          "name": "deepseek-v4.1-flash"
        },
        "qwen3-8b": {
          "name": "qwen3-8b"
        }
      }
    }
  }
}

With this config you can then use the harness as such:

Add the above json config to ~/.config/opencode/opencode.jsonc.
Run chepstow token and copy the output
Run opencode and execute the /connect command
Select isambard, paste the token, then select your model

API tokens are limited to 24 hours

API token lifetimes are limited to 24 hours. You must re-run chepstow auth, chepstow token and update the opencode credentials after expiry.