Client integrationsBeginnersUpdated

Set up Exo Harness

This guide covers Exo Harness (exoharness/exo). The separate EXO project distributes local inference across devices.

The example uses GPT-6.1 Sol (gpt-6.1-sol). Get an I Code Easy API key from API keys. Choose the Standard endpoint for GPT in Endpoint selection.

Installation

Install Rust, Cargo, and pnpm, then build Exo from its official repository. The first run may download the local sandbox runtime and image.

bash
git clone https://github.com/exoharness/exo.git
cd exo
pnpm install --frozen-lockfile
cargo build -p exo
export PATH="$PWD/target/debug:$PATH"

Exo Harness

These steps use a local, single-user setup. Set the API key in your terminal and import it into the local vault with access limited to the I Code Easy API origin.

bash
export ICODEEASY_API_KEY='YOUR_API_KEY'
exo vault secret create global icodeeasy \
  --token-env ICODEEASY_API_KEY \
  --allow-origin https://api.icodeeasy.cc

Create icodeeasy-agent.md. This definition uses the Codex harness and connects to I Code Easy over HTTP Responses.

markdown
---
harness: codex
model:
  name: gpt-6.1-sol
  credential: icodeeasy
  base_url: https://api.icodeeasy.cc/v1
---

Answer in a short sentence.

To use https://jp.icodeeasy.cc or https://sg.icodeeasy.cc, keep /v1 in base_url and authorize the backup origin with exo vault secret update global icodeeasy and --allow-origin. Repeat --allow-origin to retain the original origin as well.

Verify your first request

Run the command below to verify the connection. Omit --prompt for an interactive session.

bash
exo run icodeeasy-agent.md --prompt "Reply with a short greeting."

After the request completes, check the model and status in Usage logs. For a 401 error, check the key. For a model or permission error, check the model ID and selected endpoint.

Official client documentation