FIM Completion
FIM uses the prefix before the cursor and suffix after it. Quality and latency depend on the model, context, compatibility and network. Review code before accepting it.
Request Path
/api/fim/complete accepts path, language, prefix, suffix and generation parameters. Nonempty request-level api_url, api_key and model select the request provider; otherwise it uses the global FIM service. A chat endpoint does not necessarily support native FIM.
The HTTP entry does not wire persistent completion-cache lookup or writeback. It does not guarantee zero Token usage on cache hits or fixed latency. latency_ms measures that request.
Hosted Completion Quota
Current server daily free quotas are 500 requests for Free and 5000 for Pro and above. Beyond the daily quota, model usage consumes credits; insufficient credits may reject the request. Quota is device-scoped, so changing accounts on one device does not renew it. Device identity is required, with at most 3 distinct device identities per account per day. Running service configuration, account state and responses determine actual availability.
BYOK follows the selected channel and provider billing without consuming platform credits. Completion sends necessary code context. Never put keys in logs or project files.