Models & Inference

dsh-vision — Vision bridge: view_image tool over any OpenAI-compatible VLM (Zhipu free tier by default).
dsh-advisor — Second model passively reviews each turn and injects notes.
dsh-llm-fallbacks — Role-based LLM retry/fallback strategy.
dsh-pi-adapter — ExtensionAPI bridge for pi.
dsh-a2a — Agent2Agent mesh.
dsh-acp — Client-neutral ACP adapter.
deepseek-harness-acp — ACP profile plugin and standalone server that exposes the full DSH agent to Zed and other ACP clients while reusing DSH credentials, sessions, and MCP configuration.
dsh-mnemon — Mnemonic layer.
dsh-slice-agent-loop — Drop-in agent loop with bounded-slice context engine (cordis).
savemoneybenchmark — Cost-reduction benchmark (examples + skills).
dsh-harness-mcp-server — MCP server exposing Harness agent: any MCP client (e.g. Hermes) drives Harness as its 'arms'.
dsh-subagent-tools — Per-call model / provider / persona / toolFilter overrides for subagent delegation, @preset: references, provider/model composite ids (bundle, no patched files).
dsh-subagent-cwd — dsh-subagent-tools plus a per-call cwd for subagents, with the two in-process provider patches it requires.

See also#