AI Agent Hub
Back to plugins
🤖

dsh-plugin-devin-bridge

Model Inference Updated 2026.08.20

Run the following command in DeepSeek Harness:

dsh plugin install Arborsm/dsh-plugin-devin-bridge

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install Arborsm/dsh-plugin-devin-bridge in a DeepSeek Harness terminal to install this plugin; the source code lives at https://github.com/Arborsm/dsh-plugin-devin-bridge .

About this plugin

dsh's agent loop speaks its own LlmAdapter protocol, while Devin-hosted models (glm-5.2, swe-1.7, and beyond) sit behind a Connect RPC + protobuf boundary. dsh-plugin-devin-bridge fills that gap: it registers a devin provider route, translates dsh's GenerateOptions into Devin's GetChatMessageRequest, and decodes each protobuf response frame back into dsh's StreamChunk stream. The agent loop never knows anything changed.

The plugin covers the full response surface. Text deltas stream in, reasoning blocks map cleanly, tool-call deltas convert to dsh tool blocks, image inputs flow through the attachment system, and token usage is reported upstream. The model catalog is fetched live from Devin with a five-minute TTL cache, falling back to a static list on network hiccups so the selector never goes empty. Token resolution is three-tiered in priority—settings panel, plugin config, then the local credentials file written by devin auth login—so the plugin is ready to go the moment a Devin CLI session exists.

If you already run dsh agents, hold a Devin account, and want to steer your agent loop straight at any Devin-hosted model without hand-rolling RPC glue, this is the drop-in piece. Install it, pick devin in the model selector, and the loop handles the rest.

Use Cases

  • Route dsh agent tasks straight to Devin-hosted models such as glm-5.2 and swe-1.7
  • Tune model selection, proxy, and other settings at runtime via the dsh settings panel
  • Leverage Devin tool-call and reasoning blocks to strengthen agent workflows

Best For

  • dsh agent developers who also maintain a Devin account
  • Integration engineers who want to avoid hand-rolling Connect RPC glue
  • Production ops teams that need multi-model fallback and auto-detected credentials