dsh-agnes-paseo

If you operate both DeepSeek Harness (dsh) and Paseo in your daily workflow, wiring up a new model gateway typically means hand-editing the LLM adapter, writing an ACP bridge from scratch, and injecting a provider entry by hand—any single misstep and the integration breaks. dsh-agnes-paseo collapses all three steps into one configuration-only install with no build step and no runtime dependencies.

On the dsh side, a lightweight bundle patch declares the Agnes OpenAI-compatible endpoint on the llm-pi-ai adapter and switches the default model to agnes-2.5-flash. On the Paseo side, a zero-dependency ACP bridge script wraps dsh --profile headless as a standard ACP agent; the setup routine writes the provider entry into the Paseo config with an automatic backup of the original file. Three model tiers are available—agnes-2.5-flash for speed, agnes-2.5-pro for depth, and agnes-2.0-flash as the previous generation—all reasoning models defaulting to reasoning_effort low to balance throughput and quality.

Built for developers who already run a dsh + Paseo pipeline and want to layer Agnes reasoning models without pulling in extra build tooling or runtime dependencies. Removing the plugin instantly restores dsh's original default model; the Paseo provider entry is a one-line manual deletion, keeping the rollback path short and painless.