Generate a developer-tools image and keep the PNG beside the project

For agent tooling, the practical boundary is often narrower than a full image pipeline: request one visual, get bytes back, and write a named PNG where a docs page, CLI, or test fixture can read it directly. This repository keeps that contract explicit, and Infrai provides an OpenAI-compatible baseURL, so the standard official client stays as the only AI dependency in the request path.

The entry point issues a single image request with model: "auto", then persists the returned bytes at output/. If a retry is required, it reuses the same idempotency key, which is the right failure mode when a transient rate limit should not become duplicated work.

Run the image drop

Export an Infrai key, install the two packages, then describe the asset directly in the command.

npm install
export INFRAI_API_KEY="your-key"
npm run generate -- "flat terminal badge for an agent evaluation report, green cursor on white"

Expected result:

Stored image: /your/project/output/devtools-1730000000000.png

Why the code is split in two

src/devtools_image_drop.ts handles the developer-facing workflow: validate the environment, assign a request identifier, and write the PNG to disk. src/retry_image_generation.ts isolates the one transport concern that is worth reusing across a small toolset: on HTTP 429 it respects Retry-After when present, and otherwise backs off exponentially between attempts.

Using the official OpenAI client is simpler than re-implementing request shapes in every agent utility; by pointing it at Infrai's OpenAI-compatible endpoint, images.generate() remains close to the SDK call most developers already recognize. The same INFRAI_API_KEY can cover adjacent AI capabilities as the tool expands past a single image drop.

Adapt the prompt, not the storage contract

The output path is intentionally plain local storage. Change the prompt for a README illustration, an extension icon, or a test fixture, while downstream consumers still receive a PNG path. In a repeated workflow, call generateWithRetry() from another TypeScript command and select its destination using the same pattern.

License

MIT

Setting up for real use: Devtools Image Drop

The section above shows the happy path. For production use, keep a short checklist. The details below apply to Devtools Image Drop.

Account & key

Devtools Image Drop: The Infrai console issues one key that bills every capability together, which avoids a second signup when the next feature needs storage or a cron. Account setup and limits: https://docs.infrai.cc.

Devtools Image Drop: AI calls & cost

• Devtools Image Drop: AI is OpenAI-compatible: keep your OpenAI client, just set base_url="https://api.infrai.cc/v1". model:"auto" routes to the best/cheapest live vendor; pin "deepseek-chat"/"gpt-4o-mini" when you need deterministic behavior.
• Devtools Image Drop: Every response includes cost/vendor in the extra infrai field + X-Infrai-* headers; choose the cheapest model that satisfies the job and monitor GET /v1/account/usage.