@alisio/plugin-openai-compatible 0.1.0-alpha.8 → 0.1.0

package/README.md

@@ -1,7 +1,57 @@
 # @alisio/plugin-openai-compatible
 
-Built-in OpenAI-compatible provider for Alisio. It supports Chat Completions and Responses APIs,
-model discovery, tool calls, usage metadata, reasoning deltas, and image attachments.
+**The built-in OpenAI-compatible model provider for [Alisio](https://github.com/GustavoGutierrez/alisio).**
+Point it at any Chat Completions or Responses endpoint — OpenAI, a gateway, or a local server such
+as llama.cpp — and use it from `/connect`.
 
-Configuration and credentials are supplied by the Alisio provider registry; the plugin does not
-persist secrets itself.
+## What it is
+
+A `model-provider` plugin that registers the `openai-compatible` provider with the Alisio host. It
+ships as a built-in of the `alisio` CLI and is enabled by default. Configuration and credentials
+are supplied by the Alisio provider registry; the plugin does not persist secrets itself.
+
+## Installation
+
+Preinstalled in the `alisio` CLI. For an embedded host, register it through `createApplication`'s
+`builtins` list.
+
+## Using it: `/connect`
+
+1. Run `alisio`, then `/connect` and choose **OpenAI compatible**.
+2. Fields: **Base URL** (default `https://api.openai.com/v1`), **API key** (secret, optional),
+   **API key environment variable** (default `OPENAI_API_KEY`), **API mode** (`chat` or
+   `responses`), **Authentication** (`bearer` or `none`), **Token parameter** (`max_tokens`,
+   `max_completion_tokens` or `omit`), **Stream usage** (boolean).
+3. Pick a model from the discovered catalog — or enter one manually when the endpoint exposes no
+   `GET /models` catalog (`/connect` then asks for an optional context window in tokens).
+
+The key is read from the stored credential first, then from `process.env[apiKeyEnv]`:
+
+```sh
+export OPENAI_API_KEY=...   # never put the key in config files, profiles or READMEs
+```
+
+## Features
+
+- Model discovery via `GET /models`; per-model context windows (overridable per profile).
+- Chat Completions and Responses modes; bearer or no authentication; configurable token parameter.
+- Tool calls, usage metadata, reasoning deltas and image attachments.
+- Provider-native web search available when the API mode is `responses`.
+- Legacy `provider` configuration (`baseURL`, `apiKeyEnv`, `model`, `apiMode`, `auth`,
+  `tokenParameter`, `streamUsage`) keeps working for startup and headless runs.
+
+While registered, the provider also appears in `/model` and `alisio doctor`.
+
+## Docs
+
+[Configuration](https://gustavogutierrez.github.io/alisio/configuration) ·
+[Plugins](https://gustavogutierrez.github.io/alisio/plugins).
+
+## Requirements
+
+Node.js **>= 22.16** or Bun **>= 1.4.2**.
+
+## License
+
+MIT. Maintained by Gustavo Gutiérrez Mercado. Source: <https://github.com/GustavoGutierrez/alisio> ·
+npm: <https://www.npmjs.com/settings/alisio/packages>.

package/dist/provider.js

@@ -135,10 +135,11 @@ export class OpenAICompatibleProvider {
         if (finish !== "stop" && finish !== "tool_calls" && finish !== "length")
             throw new Error(`Provider response incomplete: ${finish ?? "stream ended"}`);
         const completed = [...calls.entries()].sort(([a], [b]) => a - b).map(([, c]) => c);
-        if (completed.some((c) => !c.id || !c.name))
+        // A response cut by the output limit is reported as it is (`truncated: true`), even when it
+        // has no text or its tool calls are partial: the host decides whether to recover, and never
+        // executes a call whose arguments are incomplete.
+        if (finish !== "length" && completed.some((c) => !c.id || !c.name))
             throw new Error("Incomplete tool call");
-        if (finish === "length" && !text.trim())
-            throw new Error("Provider response cut off by max output tokens before any usable content; raise limits.maxOutputTokens (/settings → Agent max output tokens)");
         yield {
             type: "completed",
             message: {
@@ -215,10 +216,7 @@ export class OpenAICompatibleProvider {
                     .filter((x) => x.type === "output_text")
                     .map((x) => x.text)
                     .join("");
-                if (!text.trim())
-                    throw new Error("Provider response cut off by max output tokens before any usable content; raise limits.maxOutputTokens (/settings → Agent max output tokens)");
-                if (calls.some((c) => !c.id || !c.name))
-                    throw new Error("Incomplete tool call");
+                // Reported as truncated even when empty or with partial calls: the host recovers.
                 yield {
                     type: "completed",
                     message: { role: "assistant", text, calls, providerData: r.output, truncated: true },

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@alisio/plugin-openai-compatible",
-  "version": "0.1.0-alpha.8",
+  "version": "0.1.0",
   "description": "Built-in OpenAI-compatible model provider for Alisio.",
   "license": "MIT",
   "type": "module",
@@ -24,10 +24,10 @@
     "openai": "7.23.0"
   },
   "peerDependencies": {
-    "@alisio/sdk": "^0.1.0-alpha.6"
+    "@alisio/sdk": "^0.1.0"
   },
   "devDependencies": {
-    "@alisio/sdk": "0.1.0-alpha.6"
+    "@alisio/sdk": "0.1.0"
   },
   "publishConfig": {
     "access": "public",