mohdel 3.1.0 → 3.2.0

package/README.md

@@ -1,6 +1,6 @@
 # Mohdel
 
-Self-hosted LLM gateway and SDK for Node — think LiteLLM, for the JS world. One `answer()` call for 13 providers or local inference; swap models by changing one string; get real per-call USD cost back on every result, with OpenTelemetry built in and process isolation when you need it. Your keys, your infra, no SaaS proxy in the path.
+Self-hosted LLM gateway and SDK for Node — think LiteLLM, for the JS world. One `answer()` call for 14 providers or local inference; swap models by changing one string; get real per-call USD cost back on every result, with OpenTelemetry built in and process isolation when you need it. Your keys, your infra, no SaaS proxy in the path.
 
 ```bash
 npm install -g mohdel
@@ -20,7 +20,7 @@ curate novita` writes complete, priced entries on its own, and setup counts the
 models that cost nothing and offers to add all of them in one keystroke. A
 working catalog without a pricing page or a brief.
 
-Providers: Anthropic, OpenAI, Gemini, Mistral, Groq, xAI, Cerebras, Fireworks, DeepSeek, Qwen Cloud, Xiaomi, OpenRouter, Novita. Node 22+, ES modules.
+Providers: Anthropic, OpenAI, Gemini, Mistral, Groq, xAI, Cerebras, Fireworks, DeepSeek, Qwen Cloud, Xiaomi, Meta Model API, OpenRouter, Novita. Node 22+, ES modules.
 
 Mohdel runs the inference layer of production stacks, among them [docAnalyzer](https://docanalyzer.ai), a document analysis and chat platform serving hundreds of thousands of users.
 
@@ -422,6 +422,7 @@ OPENROUTER_API_SK=sk-or-...
 NOVITA_API_SK=...
 QWEN_API_SK=sk-...
 XIAOMI_API_SK=...
+META_API_SK=...
 COHERE_API_SK=...
 MOHDEL_LOCAL_API_SK=...
 ```
@@ -461,6 +462,7 @@ What each provider supports through mohdel's unified interface:
 | Mistral | Yes | Yes | Yes | No | No | `tool_choice: "any"` = required |
 | Qwen Cloud | Yes | Yes | No | No | Yes (`enable_thinking` + `thinking_budget`) | Alibaba DashScope intl; hybrid models think by default — effort `none` sends explicit off |
 | Xiaomi | Yes | Yes | Yes | No | Auto | MiMo; shared chat-completions path, `reasoning_content` captured |
+| Meta | Yes | Yes | Yes | No | Yes (`reasoning.effort`) | OpenAI Responses API over `api.meta.ai/v1`, sent with `store: false`; Muse Spark reasoning cannot be turned off |
 | OpenRouter | Yes | Yes | Yes | No | Varies | Meta-provider; `providerOptions.openrouter` for routing prefs |
 | Local | Yes | Yes | Yes | No | No | Any OpenAI-compatible server; endpoint is the catalog entry's `baseURL` |
 | Cohere | n/a | n/a | n/a | n/a | n/a | Embeddings only: no chat models reach mohdel through it |

package/js/session/adapters/_registry.js

@@ -20,6 +20,7 @@ export const ADAPTER_NAMES = Object.freeze([
   'gemini',
   'groq',
   'local',
+  'meta',
   'mistral',
   'novita',
   'openai',

package/js/session/adapters/index.js

@@ -20,6 +20,7 @@ import { fireworks } from './fireworks.js'
 import { gemini } from './gemini.js'
 import { groq } from './groq.js'
 import { local } from './local.js'
+import { meta } from './meta.js'
 import { mistral } from './mistral.js'
 import { novita } from './novita.js'
 import { openai } from './openai.js'
@@ -38,6 +39,7 @@ export const adapters = Object.freeze({
   gemini,
   groq,
   local,
+  meta,
   mistral,
   novita,
   openai,

package/js/session/adapters/meta.js

@@ -0,0 +1,29 @@
+/**
+ * Meta Model API adapter — OpenAI Responses API over api.meta.ai/v1.
+ * Delegates to the `openai` adapter with a baseURL-configured client;
+ * the openai adapter branches on `providerOf(envelope.model)` for the
+ * fields that differ between vendors.
+ *
+ * @module session/adapters/meta
+ */
+
+import OpenAI from 'openai'
+
+import { openai } from './openai.js'
+import { streamingDispatcher } from './_dispatcher.js'
+
+const BASE_URL = 'https://api.meta.ai/v1'
+
+/**
+ * @param {import('#core/envelope.js').CallEnvelope} envelope
+ * @param {{client?: any, signal?: AbortSignal}} [deps]
+ * @returns {AsyncGenerator<import('#core/events.js').Event>}
+ */
+export async function * meta (envelope, deps = {}) {
+  const client = deps.client ?? new OpenAI({
+    apiKey: envelope.auth.key,
+    baseURL: BASE_URL,
+    fetchOptions: { dispatcher: streamingDispatcher() }
+  })
+  yield * openai(envelope, { ...deps, client })
+}

package/js/session/adapters/openai.js

@@ -360,6 +360,9 @@ function buildRequest (envelope, input, instructions) {
 
   if (envelope.speed) request.service_tier = envelope.speed
 
+  // Meta's Responses API stores every prompt and response unless told not to.
+  if (provider === 'meta') request.store = false
+
   return request
 }
 

package/js/session/run.js

@@ -140,6 +140,16 @@ export async function * run (envelope, {
     return
   }
 
+  if (envelope.outputEffort) {
+    const effortErr = effortError(key, envelope.outputEffort, spec)
+    if (effortErr) {
+      log.warn({ provider, effort: envelope.outputEffort }, '[mohdel:answer] unsupported output effort')
+      endSpanError(span, new Error(effortErr.error.message))
+      yield effortErr
+      return
+    }
+  }
+
   if (envelope.speed) {
     const speedErr = speedError(key, envelope.speed, spec, provider, adapter)
     if (speedErr) {
@@ -337,29 +347,40 @@ export function normalizeModelId (envelope, resolveSpec) {
     return { envelope: next, key: base, spec: baseSpec }
   }
 
-  if (!baseSpec.thinkingEffortLevels) {
-    return {
-      ...unresolved,
-      error: errorEvent(
-        `Model '${base}' does not support output effort (no thinkingEffortLevels). Cannot use ':${effort}' suffix.`,
-        'SESSION_INVALID_OUTPUT_EFFORT'
-      )
-    }
-  }
-  if (effort !== 'none' && !baseSpec.thinkingEffortLevels[effort]) {
-    return {
-      ...unresolved,
-      error: errorEvent(
-        `Model '${base}' does not support output effort level '${effort}'. Available: ${Object.keys(baseSpec.thinkingEffortLevels).join(', ')}`,
-        'SESSION_INVALID_OUTPUT_EFFORT'
-      )
-    }
-  }
+  const effortErr = effortError(base, effort, baseSpec)
+  if (effortErr) return { ...unresolved, error: effortErr }
 
   next.outputEffort = effort
   return { envelope: next, key: base, spec: baseSpec }
 }
 
+/**
+ * An effort level is valid only when the entry's `thinkingEffortLevels`
+ * declares it — `none` included, since some models cannot turn
+ * thinking off.
+ *
+ * @param {string} key
+ * @param {string} effort
+ * @param {any} spec
+ * @returns {import('#core/events.js').ErrorEvent | undefined}
+ */
+export function effortError (key, effort, spec) {
+  const levels = spec.thinkingEffortLevels
+  if (!levels) {
+    return errorEvent(
+      `Model '${key}' does not support output effort (no thinkingEffortLevels). Cannot use effort '${effort}'.`,
+      'SESSION_INVALID_OUTPUT_EFFORT'
+    )
+  }
+  if (!Object.hasOwn(levels, effort)) {
+    return errorEvent(
+      `Model '${key}' does not support output effort level '${effort}'. Available: ${Object.keys(levels).join(', ')}`,
+      'SESSION_INVALID_OUTPUT_EFFORT'
+    )
+  }
+  return undefined
+}
+
 /**
  * A `cache` marker is honoured on `text` parts only, with a TTL the
  * adapters know. The adapters would ignore any other marker, so it is

package/package.json

@@ -1,12 +1,12 @@
 {
   "name": "mohdel",
-  "version": "3.1.0",
+  "version": "3.2.0",
   "license": "MIT",
   "author": {
     "name": "Christophe Le Bars",
     "email": "clb@toort.net"
   },
-  "description": "Self-hosted LLM gateway and SDK for Node — a LiteLLM-style unified API for 13 providers (Anthropic, OpenAI, Gemini, Mistral, Groq, xAI, DeepSeek, OpenRouter, …) with per-call USD cost tracking, streaming, tool calls, vision, speech-to-text, and built-in OpenTelemetry. Run in-process, or behind the process-isolated thin-gate for fault containment.",
+  "description": "Self-hosted LLM gateway and SDK for Node — a LiteLLM-style unified API for 14 providers (Anthropic, OpenAI, Gemini, Mistral, Groq, xAI, DeepSeek, OpenRouter, …) with per-call USD cost tracking, streaming, tool calls, vision, speech-to-text, and built-in OpenTelemetry. Run in-process, or behind the process-isolated thin-gate for fault containment.",
   "type": "module",
   "keywords": [
     "llm",
@@ -135,11 +135,11 @@
   "optionalDependencies": {
     "@opentelemetry/exporter-trace-otlp-grpc": "^0.222.0",
     "@opentelemetry/sdk-node": "^0.222.0",
-    "chalk": "^6.0.0",
-    "mohdel-thin-gate-linux-x64-gnu": "3.1.0"
+    "chalk": "^6.0.1",
+    "mohdel-thin-gate-linux-x64-gnu": "3.2.0"
   },
   "dependencies": {
-    "@anthropic-ai/sdk": "^0.128.0",
+    "@anthropic-ai/sdk": "^0.129.0",
     "@cerebras/cerebras_cloud_sdk": "^1.91.0",
     "@clack/prompts": "^1.8.1",
     "@google/genai": "^2.24.0",
@@ -154,7 +154,7 @@
   },
   "devDependencies": {
     "gpt-tokenizer": "^4.0.0",
-    "lint-staged": "^17.5.1",
+    "lint-staged": "^17.6.0",
     "release-it": "^21.0.3",
     "standard": "^17.1.2",
     "typescript": "^7.0.2",

package/src/lib/creators.js

@@ -36,7 +36,7 @@ const creators = {
     description: 'Kuaishou\'s KwaiPilot team builds KAT-Coder, a MoE coding model with strong agentic and multi-step reasoning for software engineering tasks.'
   },
   meta: {
-    prefixes: ['llama', 'code-llama'],
+    prefixes: ['llama', 'code-llama', 'muse'],
     label: 'Meta',
     logo: 'meta.svg',
     description: 'Meta stewards the Llama ecosystem with open, widely adoptable models for chat, coding, and research.'

package/src/lib/index.js

@@ -402,7 +402,7 @@ const mohdel = async ({ logger, verbosity: verbosityOpt, onSuccess, onFailure, c
             if (!modelSpec.thinkingEffortLevels) {
               throw new Error(`Model '${resolvedModelId}' does not support output effort (no thinkingEffortLevels). Cannot use ':${aliasOutputEffort}' suffix.`)
             }
-            if (aliasOutputEffort !== 'none' && !modelSpec.thinkingEffortLevels[aliasOutputEffort]) {
+            if (!Object.hasOwn(modelSpec.thinkingEffortLevels, aliasOutputEffort)) {
               throw new Error(`Model '${resolvedModelId}' does not support output effort level '${aliasOutputEffort}'. Available: ${Object.keys(modelSpec.thinkingEffortLevels).join(', ')}`)
             }
           }

package/src/lib/provider-info.js

@@ -79,6 +79,13 @@ const PROVIDER_INFO = {
     hint: 'Create an API key at novita.ai → Dashboard → API Key',
     free: false
   },
+  meta: {
+    label: 'Meta Model API',
+    description: 'Muse Spark — reasoning, long context, image, video and PDF input.',
+    url: 'https://dev.meta.ai/',
+    hint: 'Create an API key in the Meta Model API console at dev.meta.ai',
+    free: false
+  },
   xiaomi: {
     label: 'Xiaomi MiMo',
     description: 'MiMo — vision and text models.',

package/src/lib/providers.js

@@ -86,6 +86,18 @@ const providers = {
     contextSemantics: 'shared',
     outputCapStrategy: 'accept'
   },
+  meta: {
+    sdk: 'openai',
+    apiKeyEnv: 'META_API_SK',
+    baseURL: 'https://api.meta.ai/v1',
+    createConfiguration: apiKey => ({ apiKey }),
+    references: {
+      pricing: 'https://dev.meta.ai/docs/pricing-rate-limits',
+      models: 'https://dev.meta.ai/docs/models',
+      rateLimits: 'https://dev.meta.ai/docs/pricing-rate-limits'
+    },
+    outputCapStrategy: 'accept'
+  },
   cohere: {
     sdk: 'cohere',
     api: 'embeddings',