feat(provider/aimlapi): add aimlapi.com LLM support

.env.sample

@@ -18,6 +18,9 @@ LEON_ZAI_API_KEY=
 # MiniMax API key
 LEON_MINIMAX_API_KEY=
 
+# AI/ML API (aimlapi.com) API key
+LEON_AIMLAPI_API_KEY=
+
 # OpenAI API key
 LEON_OPENAI_API_KEY=
 

config.sample.yml

@@ -56,6 +56,12 @@ llm:
       api_key:
         # The value is read from LEON_MINIMAX_API_KEY in your profile .env file.
         env: LEON_MINIMAX_API_KEY
+    aimlapi:
+      # aimlapi.com OpenAI-compatible Chat Completions endpoint.
+      base_url: https://api.aimlapi.com/v1
+      api_key:
+        # The value is read from LEON_AIMLAPI_API_KEY in your profile .env file.
+        env: LEON_AIMLAPI_API_KEY
     openai:
       api_key:
         # The value is read from LEON_OPENAI_API_KEY in your profile .env file.

scripts/commit-msg.js

@@ -20,7 +20,7 @@ import { LogHelper } from '@/helpers/log-helper'
         'utf8'
       )
       const regex =
-        '(build|BREAKING|chore|ci|docs|feat|fix|perf|refactor|style|test)(\\((web app|scripts|server|agent mode|controlled mode|aurora|messaging app|built-in command|hotword|python tcp server|bridge\\/(python|nodejs)|tool\\/([\\w-]+)|skill\\/([\\w-]+)|provider\\/(llamacpp|sglang|openrouter|zai|minimax|openai|anthropic|moonshotai|huggingface|cerebras|groq|celeris)))?\\)?: .{1,50}'
+        '(build|BREAKING|chore|ci|docs|feat|fix|perf|refactor|style|test)(\\((web app|scripts|server|agent mode|controlled mode|aurora|messaging app|built-in command|hotword|python tcp server|bridge\\/(python|nodejs)|tool\\/([\\w-]+)|skill\\/([\\w-]+)|provider\\/(llamacpp|sglang|openrouter|zai|minimax|aimlapi|openai|anthropic|moonshotai|huggingface|cerebras|groq|celeris)))?\\)?: .{1,50}'
 
       if (commitMessage.match(regex) !== null) {
         LogHelper.success('Commit message validated')

server/src/config.ts

@@ -120,6 +120,12 @@ const DEFAULT_CONFIG: LeonConfig = {
           env: 'LEON_MINIMAX_API_KEY'
         }
       },
+      aimlapi: {
+        base_url: 'https://api.aimlapi.com/v1',
+        api_key: {
+          env: 'LEON_AIMLAPI_API_KEY'
+        }
+      },
       openai: {
         api_key: {
           env: 'LEON_OPENAI_API_KEY'

server/src/core/llm-manager/llm-model-catalog.ts

@@ -115,6 +115,29 @@ const ROUTABLE_SPEED = [
  * setup and command autocomplete remain deterministic and work offline.
  */
 export const LLM_MODEL_CATALOG: readonly LLMModelCatalogEntry[] = [
+  /**
+   * aimlapi.com is an aggregator, so the model id is the label, as for
+   * OpenRouter. Every id below was checked against the live catalog
+   * (`GET /v1/models?include=all`, type `openai/chat-completions`) and
+   * answered a real completion. The dotted spelling is deliberate: the dashed
+   * `anthropic/claude-*` variants publish only `streaming` capabilities.
+   * Reasoning is left on `auto` because the gateway rejects the
+   * vendor-specific reasoning fields Leon would otherwise attach.
+   *
+   * @see https://docs.aimlapi.com/api-references/model-database
+   */
+  { provider: LLMProviders.AIMLAPI, model: 'openai/gpt-5.6-sol', label: 'openai/gpt-5.6-sol', recommended: true, reasoning: AUTO_REASONING, speed: AUTO_SPEED },
+  /** @see https://docs.aimlapi.com/api-references/model-database */
+  { provider: LLMProviders.AIMLAPI, model: 'anthropic/claude-opus-4.8', label: 'anthropic/claude-opus-4.8', reasoning: AUTO_REASONING, speed: AUTO_SPEED },
+  /** @see https://docs.aimlapi.com/api-references/model-database */
+  { provider: LLMProviders.AIMLAPI, model: 'anthropic/claude-sonnet-4.6', label: 'anthropic/claude-sonnet-4.6', reasoning: AUTO_REASONING, speed: AUTO_SPEED },
+  /** @see https://docs.aimlapi.com/api-references/model-database */
+  { provider: LLMProviders.AIMLAPI, model: 'google/gemini-3.8-flash', label: 'google/gemini-3.8-flash', reasoning: AUTO_REASONING, speed: AUTO_SPEED },
+  /** @see https://docs.aimlapi.com/api-references/model-database */
+  { provider: LLMProviders.AIMLAPI, model: 'deepseek/deepseek-v4-pro', label: 'deepseek/deepseek-v4-pro', reasoning: AUTO_REASONING, speed: AUTO_SPEED },
+  /** @see https://docs.aimlapi.com/api-references/model-database */
+  { provider: LLMProviders.AIMLAPI, model: 'moonshot/kimi-k3', label: 'moonshot/kimi-k3', reasoning: AUTO_REASONING, speed: AUTO_SPEED },
+
   /**
    * @see https://docs.celeris.ai/models Fast diffusion model for short agentic calls.
    */

server/src/core/llm-manager/llm-provider-account-configs.ts

@@ -15,6 +15,12 @@ export interface LLMProviderAccountConfig {
  */
 export const LLM_PROVIDER_ACCOUNT_CONFIGS: ReadonlyArray<LLMProviderAccountConfig> =
   Object.freeze([
+    {
+      label: 'aimlapi.com',
+      value: LLMProviders.AIMLAPI,
+      apiKeyEnv: 'LEON_AIMLAPI_API_KEY',
+      apiKeyURL: 'https://aimlapi.com/app/keys'
+    },
     {
       label: 'OpenRouter',
       value: LLMProviders.OpenRouter,

server/src/core/llm-manager/llm-provider.ts

@@ -82,6 +82,7 @@ const LLM_PROVIDERS_MAP = {
   [LLMProviders.OpenRouter]: 'openrouter-llm-provider',
   [LLMProviders.ZAI]: 'z-ai-llm-provider',
   [LLMProviders.MiniMax]: 'minimax-llm-provider',
+  [LLMProviders.AIMLAPI]: 'aimlapi-llm-provider',
   [LLMProviders.OpenAI]: 'openai-llm-provider',
   [LLMProviders.Anthropic]: 'anthropic-llm-provider',
   [LLMProviders.MoonshotAI]: 'moonshotai-llm-provider',
@@ -2674,6 +2675,7 @@ export default class LLMProvider {
           LLMProviders.SGLang,
           LLMProviders.ZAI,
           LLMProviders.MiniMax,
+          LLMProviders.AIMLAPI,
           LLMProviders.Anthropic,
           LLMProviders.MoonshotAI,
           LLMProviders.Cerebras,

server/src/core/llm-manager/llm-providers/aimlapi-llm-provider.ts

@@ -0,0 +1,80 @@
+import AISDKRemoteLLMProvider from '@/core/llm-manager/llm-providers/ai-sdk-remote-llm-provider'
+import type { ResolvedLLMTarget } from '@/core/llm-manager/llm-routing'
+import { CONFIG_MANAGER } from '@/config'
+
+const DEFAULT_BASE_URL = 'https://api.aimlapi.com/v1'
+const OFFICIAL_API_HOSTNAME = 'api.aimlapi.com'
+
+/**
+ * Attribution sent with AI/ML API requests. `HTTP-Referer` and `X-Title`
+ * follow the OpenRouter convention and identify Leon, the calling project,
+ * never the gateway.
+ */
+const AIMLAPI_ATTRIBUTION_HEADERS: Readonly<Record<string, string>> =
+  Object.freeze({
+    'HTTP-Referer': 'https://github.com/leon-ai/leon',
+    'X-Title': 'Leon',
+    'X-AIMLAPI-Source': 'agent/leon',
+    'X-AIMLAPI-Partner-ID': 'part_lcAMsJBHJpF6eW4JFtT3pJfW'
+  })
+
+function resolveBaseURL(): string {
+  return CONFIG_MANAGER.getProviderBaseURL('aimlapi') || DEFAULT_BASE_URL
+}
+
+/**
+ * The Base URL is user-configurable, so attribution is scoped to the official
+ * host. A self-hosted gateway or a third-party proxy fronting the same schema
+ * must not receive Leon's AI/ML API attribution, and a new object is returned
+ * on every call so the shared constant can never be mutated.
+ */
+export function buildAIMLAPIHeaders(baseURL: string): Record<string, string> {
+  let hostname: string
+
+  try {
+    hostname = new URL(baseURL).hostname
+  } catch {
+    return {}
+  }
+
+  if (hostname !== OFFICIAL_API_HOSTNAME) {
+    return {}
+  }
+
+  return { ...AIMLAPI_ATTRIBUTION_HEADERS }
+}
+
+/**
+ * AI/ML API is an aggregator exposing many upstream vendors behind a single
+ * OpenAI-compatible Chat Completions schema.
+ *
+ * Its validator rejects a `null` value for optional request fields
+ * (`temperature`, `top_p`, `seed`, `reasoning_effort`, `tools`, `tool_choice`,
+ * `response_format`, `max_tokens`, … all answer HTTP 400), where the upstream
+ * OpenAI API accepts them. Unset parameters must therefore be omitted from the
+ * request body rather than sent as `null`, which is what the base class does
+ * and what `test/agent/unit/aimlapi-llm-provider.spec.ts` guards.
+ *
+ * Vendor-specific reasoning extension fields are not forwarded either: the
+ * field that controls reasoning differs per upstream vendor, so the empty
+ * provider-options builder stops Leon's generic compatible adapter from
+ * attaching one that this gateway would reject.
+ *
+ * @see https://docs.aimlapi.com/api-references/text-models-llm
+ */
+export default class AIMLAPILLMProvider extends AISDKRemoteLLMProvider {
+  constructor(target: ResolvedLLMTarget) {
+    const baseURL = resolveBaseURL()
+
+    super({
+      name: 'AI/ML API LLM Provider',
+      providerName: 'aimlapi',
+      apiKeyEnv: 'LEON_AIMLAPI_API_KEY',
+      model: target.model,
+      baseURL,
+      flavor: 'openai-compatible',
+      headers: () => buildAIMLAPIHeaders(baseURL),
+      buildProviderOptions: () => ({})
+    })
+  }
+}

server/src/core/llm-manager/types.ts

@@ -26,6 +26,7 @@ export enum LLMProviders {
   OpenRouter = 'openrouter',
   ZAI = 'zai',
   MiniMax = 'minimax',
+  AIMLAPI = 'aimlapi',
   OpenAI = 'openai',
   Anthropic = 'anthropic',
   MoonshotAI = 'moonshotai',

server/src/core/session-manager/index.ts

@@ -40,6 +40,7 @@ const TITLE_REASONING_MODE_OFF_PROVIDERS = [
   LLMProviders.OpenRouter,
   LLMProviders.ZAI,
   LLMProviders.MiniMax,
+  LLMProviders.AIMLAPI,
   LLMProviders.Anthropic,
   LLMProviders.MoonshotAI,
   LLMProviders.HuggingFace,

server/src/schemas/core-schemas.ts

@@ -94,6 +94,7 @@ export const configSchemaObject = strictObject({
       openrouter: llmProvider,
       zai: llmProvider,
       minimax: llmProviderWithBaseURL,
+      aimlapi: llmProviderWithBaseURL,
       openai: llmProvider,
       anthropic: llmProvider,
       moonshotai: llmProvider,

test/agent/e2e/provider-matrix.ts

@@ -46,6 +46,14 @@ export const PROVIDER_MATRIX = [
     requiredEnv: 'LEON_MINIMAX_API_KEY',
     llmTarget: 'minimax/MiniMax-M3',
     reasoning: 'none'
+  },
+  {
+    provider: 'aimlapi',
+    requiredEnv: 'LEON_AIMLAPI_API_KEY',
+    llmTarget: 'aimlapi/openai/gpt-5.6-sol',
+    // The gateway rejects the vendor reasoning fields Leon would attach, so
+    // the curated catalog exposes auto only.
+    reasoning: null
   }
 ] as const