Releases · Zn-Dk/dsh-llm-stepfun

StepFun (阶跃星辰) model access for DeepSeek Harness.

stepfun provider route backed by an OpenAI-compatible Chat Completions adapter (SSE streaming, tool calls, image input, reasoning blocks, prompt-cache-aware usage mapping, retry policy, idle-stream watchdog).
Step Plan endpoint by default (https://api.stepfun.com/step_plan/v1), pay-as-you-go one click away.
Model catalog: step-5-preview (1M context), step-3.7-flash, step-3.5-flash, step-3.5-flash-2603, stepaudio-2.5-chat (pins modalities: ["text"]).
Settings namespace llm-stepfun plus a web client login orb/dialog and a Plugins settings card.
Debug log under $DSH_HOME/logs/stepfun/ (summary = structure only, full = complete body).
Bilingual UI (zh/en) following the DSH Web locale.
docs/model-features.{md,zh.md}: specs, pricing, rate-limit notes and a downstream reference for the TTS/ASR/Realtime models.

Image input failed on every request (the adapter passed undefined instead of the host attachments service).
readImage now receives the whole ImageAttachmentRef the host store verifies against.
429 bodies arrive as text/plain and are now surfaced instead of degrading to a bare HTTP 429.
Transport failures include their underlying cause; idle timeouts report TIMEOUT.
An unsupported reasoning effort is dropped instead of being sent.

Only the Chat Completions surface is integrated; TTS/ASR/Realtime are documented but not implemented.
37 unit tests cover the catalog, request serialization and SSE translation.

Install / 安装

dsh plugin --profile web add dsh-llm-stepfun

Navigation Menu