siliconflow/dsh-llm-siliconflow v0.2.0-rc.1

0.2.0-rc.1 — initial public release (VLM, reasoning, live discovery)

Repository: siliconflow/dsh-llm-siliconflow

Tag: v0.2.0-rc.1

Published: 2026-09-14T05:33:39Z

Prerelease: yes

Release notes: Initial public release of the SiliconFlow chat-completions adapter for the DeepSeek Harness LLM seam.

• Direct fetch + SSE (eventsource-parser) translating the OpenAI-compatible wire format to StreamChunk: delta reasoning_content → harness reasoning blocks, replayed on tool-call turns as hosted reasoning models require.
• VLM/image support: image blocks in user messages resolve via ctx.attachments into base64 data URL image_url parts; without the storage service images degrade to a placeholder and the request still goes out.
• Provider route siliconflow with live model discovery from GET /models?sub_type=chat (5-minute cache, endpoint order), fallback catalog otherwise.
• Interactive setup wizard dsh-siliconflow-setup: key → live discovery → default-model pick → writes $DSH_HOME/settings.yaml.
• Per-request resolution of baseURL/apiKey; no raw keys in config — the key is referenced by env name and resolved per request through the credentials seam.

Known limitation of this exact version: it links era-old dsh names (CallId) and the flat credentials layout. On dsh ≥ 0.1.2 it fails at ESM link time (does not provide an export named 'CallId') and its wizard can corrupt version-1 credentials files. Both were fixed in 0.2.0-rc.2 / 0.2.0-rc.3 — install one of those instead.

npm: 0.1.0-rc.5 → 0.1.0-rc.7 were internal pre-releases of this same line; 0.2.0-rc.1 is the first version under the 0.2 line.

Full change details: #1