itchenshi/dsh-gateway-models

A host-only plugin: it supplies the protocol, endpoint and model lists the engine cannot derive for gateway routes (OpenCode Go + Command Code), and attaches a session header to OpenCode requests.

What it does

• Declares the opencode-go route's wire protocol (api: openai-completions), so a catalog-unknown model no longer fails with needs an api.
• At startup it reads the engine's own model catalog, adds the DeepSeek V4.1 models to the opencode-go list and puts them first (entries you configured are kept, in their own order).
• Supplies https://opencode.ai/zen/go/v1 when the route has no baseURL; an address you set yourself is left untouched.
• Declares api + baseURL for the commandcode route (you supply only the key), and completes its model list from the provider's public catalog (add-only, idempotent).
• Attaches x-opencode-session to requests routed to OpenCode / OpenCode Go: a random UUID per session by default, and the internal session ID is never sent to a third party (mode: 'session-id' is an explicit opt-in).

Install

• Using DSH Ready GUI (recommended): the plugin ships inside the GUI. Open the GUI → Settings → "Third-party plugins" and tick Gateway routes, then restart the engine as prompted (it has no page half).

• Any other DSH host (dsh web, the CLI) — either route works:

  # Recommended: install straight from GitHub
  dsh plugin --profile web add github:itchenshi/dsh-gateway-models
  
  # Fallback: install this release's tarball
  dsh plugin --profile web add https://github.com/itchenshi/dsh-gateway-models/releases/download/v0.2.6/dsh-gateway-models-0.2.6.tgz
  

  Both land the full repository contents (including cordis.patch.yml); no extra configuration is needed afterwards.

Not on npm yet (registry.npmjs.org has no such package), so use one of the two routes above.

Usage

There is no page half and no window to open — after the engine restarts the plugin edits the route configuration for you. Confirm it worked in the session model selector: the first entry is DeepSeek V4.1 Flash, and it comes first under llm-pi-ai.providers.opencode-go.

The Command Code half touches settings only for routes you configured yourself (a user-layer entry, e.g. after adding the key), and recognises them by endpoint (api.commandcode.ai) — the route may be named commandcode, commandcode-goat or anything else.

Configuration

The config of the row id: opencode-go in cordis.patch.yml, all optional:

Key	Default	Meaning
providers	[opencode, opencode-go]	route names to attach the session header to; add your own route key
commandcodeProviders	none	extra Command Code route ids; only needed when a route's endpoint is not visible in the config
mode	uuid	uuid = random UUID per session (safe default); session-id = send the internal session ID
debug	false	true logs every streaming call that received the header
debugFile	none	path to append JSON lines to; only paths under $DSH_HOME/logs or the system temp dir take effect, and appending stops at 8 MiB

Compatibility

• Requires DSH >= 0.2.0-rc.2 (engine 0.2.0 replaced the settings service API; on 0.1.x the route configuration cannot be written).
• Node >= 20.
• It depends on the public contracts of the assembly layer, the settings service and the llm events; if those change, the engine explicitly reports a plugin load failure instead of failing silently.

FAQ

The V4.1 model is missing? Look for [gateway-models] active for providers [opencode, opencode-go] with mode uuid (debug=off) in the host log. If you instead see llm.discoverModels() is unavailable or could not detect the "opencode-go" model catalog, the engine's catalog is unreadable — the plugin then skips the auto-add rather than guessing.

The Command Code model list was not completed? The normal line is [gateway-models] "commandcode": N model(s) from the provider catalog (list X -> Y); if you see could not read the Command Code model catalog (...), that public catalog request failed and the existing list is left alone.

Does it read my API key? No. It only writes route configuration; the engine resolves the key itself from apiKeyEnv. Its single outbound request is GET https://api.commandcode.ai/provider/v1/models — public, no key required, no Authorization header. It registers no local HTTP route either.

Can a failed write leave half a config? No. Writes go through the settings service and re-run the engine's strict validation; a rejected write is refused as a whole, with the reason logged.

Development

node --check lib/index.js
npm test        # node tests/test.mjs: local behaviour tests (no network, no engine)

Add an entry to V4_1_MODELS to support another V4.1 model. planRouteUpdate() / withV41ModelsFirst() / withV41Defaults() / isV41() decide "what to write", "how to hoist", "how to fill gaps" and "is it a v4.1 model", and all four are unit-tested.

License

MIT