@staticeng/opencode-litellm 0.2.3 → 0.2.5

package/README.md

@@ -48,17 +48,20 @@ Shared or synchronized OpenCode configuration should use the published package n
 | `providerOverrides` | `{}` | Provider-level deep-merge overrides. |
 | `includeModes` | `chat`, `responses` | LiteLLM modes to include. |
 | `includeModels` | all included modes | Optional allowlist by model name. |
+| `includeBackends` | unset | Optional exact allowlist by `model_info.backend`, unioned with `includeModels` pins. |
 | `excludeModels` | none | Optional denylist by model name. |
 | `timeoutMs` | `30000` | Per-request discovery timeout. |
 | `strict` | `false` | Throw on fetch/mapping failures instead of warning. |
 
 The LiteLLM key must have permission to call `GET /model/info` and root `GET /model_group/info`.
 
+When `includeBackends` is present, models are candidates when `model_info.backend` exactly and case-sensitively matches an entry or their exact model name is pinned by `includeModels`. Backend values are not trimmed, normalized, or partially matched; missing and non-string values do not match. An empty `includeBackends` matches no backends, so without model pins it includes no models. `excludeModels` always wins over backend matches and pins, and candidates must still pass `includeModes` (or the default `chat` and `responses` modes). When `includeBackends` is omitted, `includeModels` retains its historical behavior: a non-empty array is a model-name allowlist, while an omitted or empty array imposes no model-name allowlist.
+
 ## Overrides
 
 LiteLLM metadata does not expose every OpenCode model field. The plugin includes exact built-in contracts for the currently approved GPT-5.4, GPT-5.5, GPT-5.6, DeepSeek V4 Flash, and Qwen3.8 aliases. Use `overrides` to customize those defaults or to add metadata for unknown models.
 
-When LiteLLM publishes `model_info.supported_reasoning_efforts`, the plugin maps that list to OpenCode variants automatically for unknown models. Known contracts use exact alias membership and win over absent, incomplete, or conflicting discovery metadata. No substring, suffix, provider namespace, or base-model inference is used.
+When LiteLLM publishes `model_info.supported_reasoning_efforts`, the plugin maps that list to OpenCode variants automatically. Known contracts contribute their approved defaults; discovered extra efforts remain available unless that exact contract explicitly closes discovery. Currently only `qwen3.8-flash-next` closes discovery so unverified efforts cannot become active. No substring, suffix, provider namespace, or base-model inference is used.
 
 ```json
 {
@@ -99,9 +102,9 @@ The built-in mode matrix is:
 | GPT-5.6 Luna / Sol / Terra | `none`, `low`, `medium`, `high`, `xhigh`, `max` | `medium` |
 | DeepSeek V4 Flash | `off`, `low`, `high`, `max` | `max` |
 | Qwen3.8 27B refusal dial | `off`, `low`, `medium`, `xhigh` | `xhigh` |
-| Qwen3.8 Flash Next | `off`, `low` | `low` |
+| Qwen3.8 Flash Next | `off`, `low`, `medium`, `xhigh` | `xhigh` |
 
-DeepSeek `off` sends canonical `none`. Qwen `off` sends only `chat_template_kwargs.enable_thinking=false`, with no reasoning effort. `ultra` is not generated by a built-in contract. For requests without an explicit named variant, the plugin supplies the row's contract default through model options; an explicit variant is merged afterward and wins. Qwen's explicit `off` cleanup removes the model's default effort before serialization. The `variantDefault` compatibility marker is UI-only and is stripped before requests for exact built-in aliases.
+DeepSeek `off` sends canonical `none`. Qwen `off` sends only `chat_template_kwargs.enable_thinking=false`, with no reasoning effort. `ultra` is not generated by a built-in contract. Flash Next activates exactly `off`, `low`, `medium`, and `xhigh`: discovered extras such as `ultra` are suppressed, V2 omits `minimal`, `high`, and `max`, and the legacy OpenCode surface disables unsupported generic choices. For requests without an explicit named variant, the plugin supplies the row's contract default through model options; an explicit variant is merged afterward and wins. Qwen's explicit `off` cleanup removes the model's default effort before serialization. The `variantDefault` compatibility marker is UI-only and is stripped before requests for exact built-in aliases.
 
 The exact dynamically discovered alias `qwen3.8-flash-next` also receives verified static metadata that overrides incomplete or conflicting LiteLLM group metadata: text and image input, text output, attachment, reasoning, tool calling, and a 262144-token context. Its output limit remains unknown and is represented by the package's `0` sentinel; pricing and other modalities are not supplied. The contract is still discovery-gated and never creates the alias when LiteLLM omits it.
 

package/dist/mapping.js

@@ -169,8 +169,16 @@ function allowedByFilters(model, options) {
         return false;
     if (isRetiredModelAlias(name))
         return false;
-    if (options.includeModels?.length && !options.includeModels.includes(name))
+    const pinned = options.includeModels?.includes(name) ?? false;
+    if (options.includeBackends !== undefined) {
+        const backend = model.model_info?.backend;
+        const backendMatched = typeof backend === "string" && options.includeBackends.includes(backend);
+        if (!backendMatched && !pinned)
+            return false;
+    }
+    else if (options.includeModels?.length && !pinned) {
         return false;
+    }
     if (options.excludeModels?.includes(name))
         return false;
     const modelMode = mode(model);
@@ -319,6 +327,8 @@ function generatedModel(model, tombstones = false) {
     if (!name)
         return undefined;
     const mappedModalities = modalities(model);
+    const discoveredVariants = supportsReasoning(model) ? reasoningVariants(model) : undefined;
+    const suppressDiscoveredVariants = modelContract(name)?.suppressDiscoveredVariants === true;
     const generated = {
         id: name,
         name,
@@ -328,7 +338,7 @@ function generatedModel(model, tombstones = false) {
         modalities: mappedModalities,
         ...(mappedModalities.input.includes("image") || mappedModalities.input.includes("pdf") ? { attachment: true } : {}),
         ...(supportsReasoning(model) ? { reasoning: true } : {}),
-        ...(supportsReasoning(model) && reasoningVariants(model) ? { variants: reasoningVariants(model) } : {}),
+        ...(discoveredVariants && !suppressDiscoveredVariants ? { variants: discoveredVariants } : {}),
         ...(model.model_info?.supports_function_calling || model.model_info?.supports_tool_choice ? { tool_call: true } : {}),
     };
     const defaults = contractDefaults(name, tombstones);
@@ -360,20 +370,26 @@ function v2Variants(model) {
         return undefined;
     return Object.entries(variants).map(([id, value]) => ({ id, body: v2VariantBody(value) }));
 }
-function contractV2Variants(name, override) {
+function contractV2Variants(name, model, override) {
     const entry = modelContract(name);
     if (!entry)
         return undefined;
     const variants = entry.modes.map((mode) => ({ id: mode.id, body: { ...mode.v2Body } }));
-    if (!isRecord(override?.variants))
-        return variants;
     const byID = new Map(variants.map((variant) => [variant.id, variant]));
-    for (const [id, value] of Object.entries(override.variants)) {
-        const current = byID.get(id);
-        byID.set(id, {
-            id,
-            body: deepMerge(current?.body ?? {}, v2VariantBody(value)),
-        });
+    if (isRecord(override?.variants)) {
+        for (const [id, value] of Object.entries(override.variants)) {
+            const current = byID.get(id);
+            byID.set(id, {
+                id,
+                body: deepMerge(current?.body ?? {}, v2VariantBody(value)),
+            });
+        }
+    }
+    if (!entry.suppressDiscoveredVariants && isRecord(model.variants)) {
+        for (const [id, value] of Object.entries(model.variants)) {
+            if (!byID.has(id))
+                byID.set(id, { id, body: v2VariantBody(value) });
+        }
     }
     return Array.from(byID.values());
 }
@@ -441,7 +457,7 @@ export function mapModelsV2(models, options = {}) {
     const legacy = mapModels(models, options);
     return Object.fromEntries(Object.entries(legacy).map(([name, model]) => {
         const mapped = v2Model(model, options.baseURL);
-        const variants = contractV2Variants(name, options.overrides?.[name]);
+        const variants = contractV2Variants(name, model, options.overrides?.[name]);
         return [name, variants ? { ...mapped, variants } : mapped];
     }));
 }

package/dist/model-contracts.js

@@ -12,7 +12,7 @@ function effortModes(ids) {
         v2Body: freezeJson({ reasoning_effort: id }),
     })));
 }
-function contract(family, aliases, modes, defaultMode, metadata) {
+function contract(family, aliases, modes, defaultMode, metadata, suppressDiscoveredVariants = false) {
     if (!modes.some((mode) => mode.id === defaultMode))
         throw new Error(`Model contract ${family} has an invalid default`);
     return Object.freeze({
@@ -22,6 +22,7 @@ function contract(family, aliases, modes, defaultMode, metadata) {
         defaultMode,
         reasoning: true,
         variantDefault: false,
+        ...(suppressDiscoveredVariants ? { suppressDiscoveredVariants: true } : {}),
         ...(metadata ? { metadata } : {}),
     });
 }
@@ -44,7 +45,6 @@ const QWEN_MODES = Object.freeze([
     Object.freeze({ id: "medium", legacy: freezeJson({ reasoningEffort: "medium" }), v2Body: freezeJson({ reasoning_effort: "medium" }) }),
     Object.freeze({ id: "xhigh", legacy: freezeJson({ reasoningEffort: "xhigh" }), v2Body: freezeJson({ reasoning_effort: "xhigh" }) }),
 ]);
-const QWEN_FLASH_NEXT_MODES = Object.freeze(QWEN_MODES.slice(0, 2));
 const QWEN_FLASH_NEXT_METADATA = Object.freeze({
     modalities: Object.freeze({
         input: Object.freeze(["text", "image"]),
@@ -69,7 +69,7 @@ export const MODEL_CONTRACTS = Object.freeze([
     contract("GPT-5.6 Terra", chatGptAliases("gpt-5.6-terra"), GPT_MAX_MODES, "medium"),
     contract("DeepSeek V4 Flash", ["deepseek-v4-flash-fp8-mtp", "deepseek-v4-flash-fp8-mtp-norefusal"], DEEPSEEK_MODES, "max"),
     contract("Qwen3.8", ["qwen3.8-27b-refusal-dial"], QWEN_MODES, "xhigh"),
-    contract("Qwen3.8", ["qwen3.8-flash-next"], QWEN_FLASH_NEXT_MODES, "low", QWEN_FLASH_NEXT_METADATA),
+    contract("Qwen3.8", ["qwen3.8-flash-next"], QWEN_MODES, "xhigh", QWEN_FLASH_NEXT_METADATA, true),
 ]);
 export const RETIRED_MODEL_ALIASES = Object.freeze([
     ...chatGptAliases("gpt-5.3-codex"),

package/dist/types.d.ts

@@ -23,6 +23,7 @@ export type ModelContract = Readonly<{
     defaultMode: string;
     reasoning: true;
     variantDefault: false;
+    suppressDiscoveredVariants?: true;
     metadata?: ModelContractMetadata;
 }>;
 export type OpenCodeModel = JsonObject & {
@@ -123,6 +124,8 @@ export type LiteLLMPluginOptions = {
     includeModes?: string[];
     excludeModels?: string[];
     includeModels?: string[];
+    /** Exact model_info.backend matches, unioned with includeModels pins. */
+    includeBackends?: string[];
     timeoutMs?: number;
     strict?: boolean;
 };
@@ -134,6 +137,7 @@ export type LiteLLMModelInfo = JsonObject & {
     model_info?: JsonObject & {
         id?: string;
         base_model?: string;
+        backend?: unknown;
         mode?: string;
         input_cost_per_token?: number | string;
         output_cost_per_token?: number | string;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@staticeng/opencode-litellm",
-  "version": "0.2.3",
+  "version": "0.2.5",
   "description": "OpenCode plugin that discovers LiteLLM models at runtime.",
   "type": "module",
   "license": "MIT",