graphify extract . --force (kimi ratelimit)

[bercedev]

I think you have to wait a second when sending a request to the API, otherwise this warning keeps popping up on the screen. Thanks. I love this project.

[safishamsi]

Thanks @bercedev — and you diagnosed it exactly right. Those are kimi/Moonshot HTTP 429s (max organization concurrency / Organization Rate limit exceeded, please try again after 1 seconds). graphify runs chunks in parallel, and the OpenAI SDK only retried twice by default — not enough for Moonshot's tight per-org limits — so the chunk gave up, logged chunk N failed, and was dropped (incomplete graph + the spam you saw).

Fixed on v8 (64c1f21, ships in the next release): the OpenAI-compatible / Azure / Anthropic clients are now built with a higher max_retries (default 6, override via GRAPHIFY_MAX_RETRIES). The SDK already backs off and honors the "try again after 1 seconds" hint — it just needed more attempts to wait out the rate window, so rate-limited chunks now succeed on retry instead of being lost.

In the meantime (and for very tight accounts), graphify extract . --max-concurrency 1 reduces the parallelism that triggers the org-concurrency limit. Thanks for the kind words and the clear screenshot!