README.md

OpenAI Activity

Activity that sends a JSON request to https://api.openai.com/v1/chat/completions and responds with the first generated response. It implements the stargazers:llm/llm WIT interface.

Prerequisites

OpenAI token is required. Navigate to API Tokens and crate a new restricted token. Make sure the resource /v1/chat/completions under Model capabilities has the write permission. The token must be accesible as OPENAI_API_KEY environment variable.

export OPENAI_API_KEY="..."

Configuration

The second argument of respond WIT function is settings, which is JSON encoded string. The corresponding Settings struct in lib.rs provides the details on the structure, and the request_should_succeed integration test shows how to set up a model, send a system message and restrict the number of tokens returned by the service. Those settings could be encoded directly in llm.wit, but were omitted for simplicity.

Running the activity

Build the activity and run Obelisk with obelisk-local.toml configuration in the root of the repository.

cargo build --release
obelisk server run --config ./obelisk-local.toml

In another terminal run the activity.

Use the following parameters:

• parameter user-prompt: "Tell me about Rust programming."
• parameter settings-json: "{\"model\": \"gpt-3.5-turbo\",\"max_tokens\": 50}"

obelisk execution submit --follow stargazers:llm/llm.respond '["Tell me about Rust programming", "{\"model\": \"gpt-3.5-turbo\", \"max_tokens\": 50}"]'

Integration testing

export TEST_OPENAI_API_KEY=...
cargo nextest run -- --ignored

To execute an ad-hoc query directly to api.openai.com using curl:

curl -X POST https://api.openai.com/v1/chat/completions \
-H "Authorization: Bearer ${OPENAI_API_KEY}" \
-H "Content-Type: application/json" \
-d '{
    "messages": [
        {
            "role": "system",
            "content": "You are a helpful assistant"
        },
        {
            "role": "user",
            "content": "Tell me about Rust programming."
        }
    ],
    "model": "gpt-3.5-turbo",
    "max_tokens": 200
}'