@mcowger/oh-my-pi-plexus 1.4.11 → 1.5.0

package/README.md

@@ -7,7 +7,6 @@ Exposes models from a self-hosted [Plexus](https://github.com/mcowger/plexus) AI
 | Package | Agent | npm |
 |---|---|---|
 | `plexus-pi` | [pi](https://github.com/earendil-works/pi) | `@mcowger/pi-plexus` |
-| `plexus-opencode` | [OpenCode](https://opencode.ai) | `@mcowger/opencode-plexus` |
 | `plexus-oh-my-pi` | [Oh My Pi](https://github.com/can1357/oh-my-pi) | `@mcowger/oh-my-pi-plexus` |
 
 ## Prerequisites
@@ -54,38 +53,6 @@ Then register the path in `~/.pi/agent/settings.json`:
 
 ---
 
-### OpenCode
-
-#### Option 1 — npm (recommended)
-
-```sh
-npm install -g @mcowger/opencode-plexus
-```
-
-Then add the plugin to your `opencode.json`:
-
-```json
-{
-  "plugins": ["@mcowger/opencode-plexus"]
-}
-```
-
-#### Option 2 — path reference
-
-```sh
-git clone https://github.com/mcowger/plexus-agent-plugins ~/code/plexus-agent-plugins
-```
-
-Then reference the built artifact in `opencode.json`:
-
-```json
-{
-  "plugins": ["~/code/plexus-agent-plugins/packages/plexus-opencode/dist/index.js"]
-}
-```
-
----
-
 ### Oh My Pi
 
 Oh My Pi is a fork of pi and is **not** wire-compatible with `plexus-pi` in every respect (different runtime packages, extension manifest field, and built-in model registry — see [AGENTS.md](AGENTS.md)), so it has its own adapter package.
@@ -161,69 +128,6 @@ Run inside Oh My Pi using the native login flow:
 
 Same prompts and `/plexus refresh` / `/plexus set-default-model` commands as pi (see above) — `plexus-oh-my-pi` mirrors the pi adapter's behavior, adjusted for Oh My Pi's own runtime packages and built-in model registry.
 
-### OpenCode
-
-Run inside OpenCode:
-
-```
-/connect
-```
-
-Select **Plexus** and enter your base URL and API key. OpenCode has no live model-discovery hook for custom providers, so the model list is seeded from the on-disk cache at startup. Run `/plexus-refresh` to force a live fetch from Plexus and rewrite the cache, then restart OpenCode to pick up the refreshed list:
-
-```
-/plexus-refresh
-```
-
-You may enter either the Plexus root URL or the `/v1` API base URL:
-
-```text
-https://plexus.example.com
-https://plexus.example.com/v1
-```
-
-The plugins normalize either form to the Plexus root URL for storage and derive `/v1` paths when calling the API.
-
-OpenCode stores the API key in its native auth store and stores the Plexus base URL as auth metadata on that connection. Existing `provider.plexus.options.plexusBaseURL` config is still honored as a fallback.
-
-`provider.plexus.options.baseURL` is also accepted as input-only compatibility config. Some external clients read OpenCode config files directly and only recognize `options.baseURL`. The plugin normalizes that value for Plexus discovery and then removes it from OpenCode's in-memory provider options, so individual models keep their own `/v1` or `/v1beta` endpoints instead of being overridden by a provider-wide URL.
-
-For an OpenChamber-compatible OpenCode config, omit `options.apiKey` and authenticate through `/connect`:
-
-```json
-{
-  "provider": {
-    "plexus": {
-      "options": {
-        "baseURL": "https://plexus.example.com/v1"
-      }
-    }
-  }
-}
-```
-
-Then run `/connect`, select **Plexus**, and authenticate against the same Plexus server. A raw unresolved `options.apiKey` value in this file can override the resolved runtime credential used by OpenChamber, including OpenCode `{env:NAME}` templates that OpenChamber does not expand when extra characters follow the template.
-
-The plugin publishes models with `/v1` endpoints before `/v1beta` models. This ordering helps OpenChamber versions that derive a provider-wide endpoint from the first runtime model; it does not change model endpoints or OpenCode routing. Gemini models still use `/v1beta` through `@ai-sdk/google`.
-
-The OpenCode plugin respects each model's `preferred_api` value and routes models through the matching SDK/API shape:
-
-- `chat_completions` / `openai-completions` → OpenAI-compatible chat completions
-- `responses` / `openai-responses` → OpenAI Responses API
-- `messages` / `anthropic-messages` → Anthropic Messages API
-- `gemini` / `google-generative-ai` → Google Gemini API
-
-You can also pre-configure via environment variables:
-
-```sh
-export PLEXUS_API_URL=https://plexus.example.com
-export PLEXUS_API_KEY=your-api-key
-```
-
-`PLEXUS_BASE_URL` is also accepted for compatibility; `PLEXUS_API_URL` wins when both are set.
-
----
-
 ## Configuration files
 
 ### pi
@@ -250,16 +154,6 @@ The API key is stored through pi's own auth storage. `PLEXUS_API_URL` or `PLEXUS
 
 Same layout as pi, rooted under `~/.omp/agent` instead of `~/.pi/agent` since Oh My Pi resolves its own agent directory.
 
-### OpenCode
-
-```
-~/.local/share/opencode/plugins/plexus/
-  models-cache.json            # last-fetched model list (startup cache)
-  models-raw.json              # raw API response (diagnostics)
-```
-
-The API key is stored through OpenCode's auth flow, and the Plexus base URL is stored as auth metadata. `PLEXUS_API_URL`, `PLEXUS_BASE_URL`, and `PLEXUS_API_KEY` can be used as environment overrides.
-
 Both adapters also accept pi-style environment interpolation in configured strings, such as `${PLEXUS_API_URL}` or `$PLEXUS_API_KEY`. This is useful when checking non-secret config into an agent config file while keeping the actual values in the environment.
 
 ---
@@ -290,19 +184,6 @@ All plugin packages support suppressing models by name or pattern so undesired m
   }
   ```
 
-- **OpenCode config**: Add `suppressModels` (or `suppress`) under `provider.plexus.options` in `opencode.json`:
-  ```json
-  {
-    "provider": {
-      "plexus": {
-        "options": {
-          "suppressModels": ["gpt-3.5*", "claude-2*"]
-        }
-      }
-    }
-  }
-  ```
-
 ---
 
 ## Package layout
@@ -331,17 +212,6 @@ packages/
       cache.ts          # model cache I/O
       log.ts            # append-only log
     package.json        # declares omp.extensions entry point
-  plexus-opencode/      # OpenCode plugin adapter
-    src/
-      plugin.ts         # Plugin export: config hook, provider hook, auth handler
-      mapper.ts         # PlexusApiModel → OpenCode ConfigModel
-      cache.ts          # model cache I/O
-      config-store.ts   # resolveConfig, readStoredAuth
-      log.ts            # logger via OpenCode SDK
-      constants.ts      # provider ID, env var names, timeouts
-      url.ts            # URL helpers (trimURL, apiBase, modelsUrl)
-      index.ts          # barrel export
-    package.json        # npm package manifest
 ```
 
 `plexus-models` has zero imports from any agent framework. Each host adapter imports it via a relative path.
@@ -355,9 +225,8 @@ The `/v1/models` endpoint returns an OpenRouter-style list. These fields drive h
 | `preferred_api` | String or array. First recognized value selects the host API dialect. |
 | `supported_parameters` | `reasoning`, `include_reasoning`, or `reasoning_effort` enables reasoning support. |
 | `architecture.input_modalities` | Enables text/image/audio/video/pdf support where the host supports it. |
-| `architecture.output_modalities` | Non-text-only output models are filtered from OpenCode chat providers. |
 | `pricing` | Converted into host per-million-token cost metadata. Plexus returns per-token prices. |
-| `pricing.tiers` | Alternate rates above `input_tokens_above`; mapped to native pi and OpenCode context pricing tiers. |
+| `pricing.tiers` | Alternate rates above `input_tokens_above`; mapped to native pi context pricing tiers. |
 | `top_provider` | Supplies context and output token limits when present. |
 | `pi_provider` / `pi_model` | Lets the pi adapter reuse built-in pi compat, headers, and thinking-level metadata. |
 | `pi_options` | pi compat overrides. These win over heuristic and built-in metadata. |
@@ -367,11 +236,7 @@ Models with a falsy `id` are skipped. Missing metadata falls back to safe defaul
 ## Adapter behavior
 
 - **pi** refreshes on session start and through `/plexus refresh`, without changing the model selected for the session. It accepts either root URLs or URLs ending in `/v1` and normalizes them before calling Plexus.
-- **OpenCode** seeds the provider from the on-disk cache (or a placeholder model) once, during config loading — OpenCode's `provider.models` hook never fires for custom providers, so there is no live discovery at startup. Run `/plexus-refresh` to force a live fetch and rewrite the cache; because OpenCode has no way to hot-reload a custom provider's model list mid-session, a restart is required afterward to see the refreshed models in the picker.
-- OpenCode models retain their upstream model ID, SDK dialect, release date, and reasoning metadata so OpenCode can generate its native GPT, Claude, Gemini, and OpenAI-compatible variants and apply its current request transforms. DeepSeek models also preserve `reasoning_content` across tool-call turns.
-- OpenCode uses a 250K-token context window when Plexus supplies no context metadata; its output fallback remains 20% of that window.
-- OpenCode moves `/v1beta` models after `/v1` models in published model order. Model endpoints and per-model SDK selection are unchanged.
-- Both adapters convert Plexus's per-token base and tier rates to the per-million-token units expected by their host.
+- The active adapters convert Plexus's per-token base and tier rates to the per-million-token units expected by their host.
 
 ## Development
 
@@ -381,6 +246,10 @@ After cloning, install dependencies to set up the pre-commit hook:
 bun install
 ```
 
-The pre-commit hook (via lefthook) rebuilds both dist artifacts automatically whenever source files change. After committing, reload/restart your agent.
+The pre-commit hook (via lefthook) rebuilds the active dist artifacts automatically whenever source files change. After committing, reload/restart your agent.
+
+## Archived adapters
+
+`packages/deprecated/plexus-opencode` is preserved for reference but is archived. It is private and excluded from builds, tests, hooks, version sync, and publishing. In my testing, OpenCode was notably slower and less token-efficient than pi and Oh My Pi.
 
 To add support for a new host agent, see [AGENTS.md](AGENTS.md).