IteraCompute Migration Kit

Install dependencies → configure your IteraCompute key → call free Qwen → see a non-empty reply.

Use the official OpenAI SDK with an explicit IteraCompute key, base URL and exact API model ID:

Setting	Value
Base URL	https://api.iteracompute.com/v1
Key variable	ITERACOMPUTE_API_KEY
Free model ID	qwen/qwen3.8-27b
Endpoint	Chat Completions

API model IDs must match the public Models endpoint. Do not add iteracompute/ to an API model ID. These examples explicitly pass ITERACOMPUTE_API_KEY; an existing OPENAI_API_KEY is never used as a fallback. The free Qwen example disables thinking to get a short visible reply; that option is specific to this model.

1. Install dependencies

Clone this repository first:

git clone https://github.com/IteraCompute/migration-kit.git
cd migration-kit

Choose Python (3.10+) or TypeScript (Node.js 22+, npm).

Python, macOS / Linux / WSL:

python3 -m venv .venv
. .venv/bin/activate
python -m pip install -r requirements.txt

Python, Windows PowerShell (activation is not required):

python -m venv .venv
.\.venv\Scripts\python.exe -m pip install -r requirements.txt

TypeScript, any platform:

npm ci

2. Configure your IteraCompute key

Create an account, verify your email or sign in with Google, then create an API key that permits qwen/qwen3.8-27b. Free access does not require a card or recharge. Save the secret shown once.

Set the variable in the same terminal you will use to run the example. Replace your-api-key with your IteraCompute key.

Bash / zsh / WSL:

export ITERACOMPUTE_API_KEY="your-api-key"

Windows PowerShell:

$env:ITERACOMPUTE_API_KEY = "your-api-key"

Keep the key on the server or in your local terminal. Do not commit it, paste it into issues, or put it in browser code. .env files are ignored but are not automatically loaded by these examples.

3. Call free Qwen

Python, macOS / Linux / WSL:

python examples/openai_python.py

Python, Windows PowerShell:

.\.venv\Scripts\python.exe examples/openai_python.py

TypeScript, any platform:

npm run quickstart

4. Check the reply

Expected stdout (npm also prints its script banner):

Hello!

Wording may vary. Success requires a non-empty choices[0].message.content and exit code 0. Missing keys, API errors and empty replies exit with code 1. Registration or HTTP 200 alone does not complete this check. Requests have a 180-second timeout and no automatic retries; free capacity may queue.

Two complete examples

Example	What you will run and verify
Text summarizer	Read a bundled meeting document, print its summary, and save a non-empty UTF-8 output file.
OpenCode coding tool	Connect free Qwen, repair a failing slugify function in a disposable project, and run four tests.

See acceptance results for the exact versions, commands, outputs and validation scope. Each example has its own complete setup and troubleshooting instructions.

Common errors

Symptom	Fix
Set ITERACOMPUTE_API_KEY	Set an actual IteraCompute key in this terminal; a placeholder, blank value or only OPENAI_API_KEY is insufficient.
ModuleNotFoundError: openai / tsx not found	Install dependencies from the repository root; use the same Python interpreter / virtual environment that ran pip. For TypeScript run npm ci.
HTTP 401	The IteraCompute key is missing, invalid or revoked. Replace it; do not use an OpenAI key.
HTTP 403	Check that the key permits qwen/qwen3.8-27b. Free Qwen does not require paid balance.
HTTP 404	Keep /v1 in the base URL and use qwen/qwen3.8-27b exactly, without an iteracompute/ prefix. Use Chat Completions.
HTTP 400	Check request options and reduce the input/output budget. Free context includes input plus requested maximum output.
HTTP 429	Check monthly allowance and TPM in the Console. Honor Retry-After; avoid parallel retries. See free limits.
Empty model reply	The sample fails even on HTTP 200. Use the supplied short prompt and Qwen thinking setting, then retry when capacity is available.
Network / timeout / HTTP 5xx	Check connectivity and provider status; retry a single request after a delay. Do not disable TLS verification.

Other models can require paid balance; qwen/qwen3.8-2.4t-a95b is a different model. Check current pricing and API documentation before switching.

Development checks

After installing both sets of dependencies:

python -m unittest discover -s tests -v

On Windows use .\.venv\Scripts\python.exe instead of python. These offline tests verify the actual SDK request body, IteraCompute authentication, missing keys, empty replies, HTTP errors and summary file writes. They use fake credentials and do not establish live API acceptance.

API docs · Integrations · Pricing CLI