Full reference

Make your first traced request

Install the CLI, select the right project, and send one request through the Stimulir gateway. Confirm the trace before moving on to evaluation.

Updated

Reviewed against Console v0.2.2. Available actions depend on project access and runtime configuration; check your installed CLI version for command differences.

Prefer a Console walkthrough?

Follow Quick start for the complete first request and trace walkthrough. In Console, select your organisation/project, choose an enabled model in Models, and use Settings → Project → API keys. The commands below provide the CLI alternative.

Install the tools

bash
uv tool install --upgrade stimulir
stimulir --version
stimulir --help

The upgrade flag also updates an existing CLI installation. See CLI installation for prerequisites, PATH troubleshooting and the pip alternative. For application code, install the SDK into that application environment with uv add stimulir. An isolated CLI install does not add the SDK to another Python environment.

Choose your workspace and project

bash
stimulir login
stimulir workspace list
stimulir workspace use <workspace-id>
stimulir project list
stimulir project use <project-id>

Choose IDs from your own account. Workspace selection applies to interactive administration; the project determines where requests and evidence belong.

Configure an application key

bash
stimulir keys create --name first-call --save
export STIMULIR_API_KEY=hyb_...
export STIMULIR_PROJECT_ID=...
export STIMULIR_MODEL=...

Copy the newly created key into the application’s secret configuration. Replace every placeholder. Do not commit the key or pass it in a shared command transcript. Run stimulir models to choose an enabled model.

Send a request

bash
curl https://api.stimulir.com/api/v1/inference/chat/completions \
  -H "Authorization: Bearer $STIMULIR_API_KEY" \
  -H "X-Project-Id: $STIMULIR_PROJECT_ID" \
  -H "Content-Type: application/json" \
  -d "{\"model\":\"$STIMULIR_MODEL\",\"messages\":[{\"role\":\"user\",\"content\":\"Say hello\"}],\"tags\":[\"first-call\"]}"

Open Traces in the same Console project. Check the model, response and recorded usage. If the request failed, keep the error code and follow request troubleshooting.

Continue with evidence

Collect a representative set of task cases before running an evaluation or training job. Data assets explains curation and snapshots; Evaluation explains controlled comparisons. Training requires a supported model family, suitable data and configured compute. It is not required to make your first call.