带记忆的 OpenRouter：统一模型 API + 智能推荐 + 记忆增强
一个 API 接入所有 LLM，自动推荐最优模型，每次调用自带记忆增强。
npm install agentkits
import { recommendModel, createChat, withBrain } from 'agentkits';
// 🎯 智能推荐：按任务 + 预算自动选最优模型
const picks = recommendModel({ task: 'coding', budget: 'medium' });
console.log(picks[0]);
// { model: 'deepseek-coder-v2', provider: 'deepseek', reason: 'Best cost/perf for coding tasks', ... }
// 🤖 统一调用：切换供应商只改一个词
const chat = createChat({ provider: 'deepseek', model: 'deepseek-chat' });
const reply = await chat.complete('用 TypeScript 写一个快排');
// 🧠 记忆增强：自动从 DeepBrain 召回上下文
const smartChat = withBrain(chat.complete.bind(chat), { brainUrl: 'http://localhost:3333' });
const response = await smartChat({ messages: [{ role: 'user', content: '继续优化昨天的方案' }] });
| 功能 | 说明 | |
|---|---|---|
| 🤖 | 统一模型 API | OpenAI / Anthropic / Google / DeepSeek / 通义 / 智谱 / Moonshot / Ollama — 一个接口搞定 |
| 🎯 | 智能推荐 | recommendModel() 按任务 + 预算 + 速度推荐最优模型 |
| 💰 | 成本估算 | estimateModelCost() 20+ 模型精确定价，按 token 算钱 |
| 🏥 | 健康检查 | checkProvider() 7 个 provider 实时可用性 + 延迟检测 |
| 🧠 | 记忆增强 | withBrain() 每次调用自动串联 DeepBrain，recall → chat → learn |
| 🎨 | Web UI | KitsUI 模型管理面板 + Playground，零配置启动 |
| 📊 | 40 个模块 | RAG / Agent / 工作流 / MCP / A2A / 视觉 / TTS / STT / 护栏 |
| 模型 | 供应商 | 输入价格 ($/1M tokens) | 输出价格 ($/1M tokens) | 上下文窗口 | 最佳场景 |
|---|---|---|---|---|---|
| gpt-4o | OpenAI | $2.50 | $10.00 | 128K | 通用、视觉、编程 |
| gpt-4o-mini | OpenAI | $0.15 | $0.60 | 128K | 简单任务、快速 |
| claude-3.5-sonnet | Anthropic | $3.00 | $15.00 | 200K | 推理、编程 |
| claude-3-haiku | Anthropic | $0.25 | $1.25 | 200K | 快速、低成本 |
| gemini-1.5-pro | $1.25 | $5.00 | 2M | 超长上下文分析 | |
| gemini-2.5-flash | $0.15 | $0.60 | 1M | 快速、低成本 | |
| deepseek-chat | DeepSeek 深度求索 | $0.14 | $0.28 | 128K | GPT-4 级质量，极致性价比 |
| deepseek-reasoner | DeepSeek 深度求索 | $0.55 | $2.19 | 128K | 深度推理 |
| qwen-turbo | 通义千问 | $0.04 | $0.08 | 128K | 快速、超低成本 |
| qwen-plus | 通义千问 | $0.11 | $0.28 | 128K | 中等质量 |
| glm-4-flash | 智谱 AI | $0.01 | $0.01 | 128K | 近乎免费 |
| glm-4-plus | 智谱 AI | $7.00 | $7.00 | 128K | 高质量中文 |
| moonshot-v1-8k | 月之暗面 (Kimi) | $0.17 | $0.17 | 8K | 中文对话 |
| grok-2 | xAI | — | — | 128K | 实时信息 |
| 本地模型 | Ollama | 免费 | 免费 | 可变 | Llama3 / Qwen2.5 / 任意 |
还支持：Yi (零一万物)、Baichuan (百川)、SiliconFlow (硅基流动)、StepFun (阶跃星辰)、MiniMax、Cohere、Fireworks、Together、Groq、Perplexity、Custom (任意 OpenAI 兼容)。
| 模型 | 供应商 | 维度 | 价格 ($/1M tokens) |
|---|---|---|---|
| text-embedding-3-small | OpenAI | 1536 | $0.02 |
| text-embedding-3-large | OpenAI | 3072 | $0.13 |
| text-embedding-004 | 768 | — | |
| text-embedding-v3 | 通义千问 | 1024 | — |
| embedding-3 | 智谱 AI | 2048 | — |
| jina-embeddings-v3 | Jina | 1024 | — |
| nomic-embed-text | Ollama | 768 | 免费 |
还支持：DeepSeek、SiliconFlow、Cohere、Voyage、Mixedbread、Fireworks、Together、Custom。
import { recommendModel, estimateModelCost } from 'agentkits';
// 按任务 + 预算推荐
const picks = recommendModel({
task: 'coding', // 'chat' | 'coding' | 'analysis' | 'embedding' | 'vision'
budget: 'low', // 'free' | 'low' | 'medium' | 'high'
speed: 'fast', // 'fast' | 'medium' | 'slow'
local: false, // true = 仅推荐 Ollama 本地模型
});
// → [{ model, provider, reason, estimatedCostPer1kTokens, speed, quality }]
// 精确成本估算
const cost = estimateModelCost('gpt-4o', 5000, 2000);
// → { cost: 0.0325, currency: 'USD', breakdown: { input: 0.0125, output: 0.02 } }
import { createChat } from 'agentkits';
const chat = createChat({ provider: 'deepseek', model: 'deepseek-chat' });
const reply = await chat.complete('你好');
// 流式输出
for await (const chunk of chat.stream('给我讲个故事')) {
process.stdout.write(chunk.content ?? '');
}
import { withBrain, recall, learn } from 'agentkits';
// 中间件模式：自动 recall → chat → learn
const smartChat = withBrain(originalChatFn, {
brainUrl: 'http://localhost:3333',
agentId: 'my-agent',
autoRecall: true,
autoLearn: true,
recallLimit: 5,
});
// 手动调用
const memories = await recall('昨天的讨论', { brainUrl: 'http://localhost:3333' });
await learn('新的知识点...', { brainUrl: 'http://localhost:3333' });
import { checkProvider } from 'agentkits';
const status = await checkProvider('openai', { apiKey: 'sk-...' });
// → { available: true, latencyMs: 234 }
// 支持的 provider: openai, anthropic, gemini, deepseek, moonshot, zhipu, ollama
import { KitsUI } from 'agentkits';
const ui = new KitsUI({ port: 4002 });
await ui.start();
// 打开 http://localhost:4002 → 模型管理面板 + Playground
// 向量化
import { createEmbedding } from 'agentkits';
const emb = createEmbedding({ provider: 'openai' });
const vector = await emb.embed('语义搜索');
// 函数调用
import { createToolChat, defineTool } from 'agentkits';
const tool = defineTool({ name: 'search', description: '搜索', parameters: {...}, handler: async (p) => '...' });
const chat = createToolChat({ provider: 'openai', tools: [tool] });
// RAG 管线
import { createRAG } from 'agentkits';
const rag = createRAG({ chatProvider: 'deepseek', embeddingProvider: 'openai' });
// Agent 循环
import { createAgent } from 'agentkits';
const agent = createAgent({ provider: 'openai', tools: [...] });
// 工作流引擎
import { createWorkflow } from 'agentkits';
// MCP 客户端
import { createMCPClient } from 'agentkits';
// A2A 协议
import { createA2AClient, createA2AServer } from 'agentkits';
// 故障转移
import { createChatWithFailover } from 'agentkits';
// 智能路由
import { createRouter, createModelRouter } from 'agentkits';
// 结构化输出、护栏、重排序、TTS、STT、图像生成...
// 完整 40 个模块见下方模块目录
| 分类 | 模块 | 引入路径 | 说明 |
|---|---|---|---|
| 核心 | 大模型对话 | agentkits/llm | 19 个供应商统一接口 |
| 向量化 | agentkits/embedding | 15 个引擎统一接口 | |
| 流式输出 | agentkits/streaming | SSE 解析、流组合、中断控制 | |
| 结构化输出 | agentkits/structured | JSON Schema 校验 | |
| 函数调用 | agentkits/function-calling | 跨供应商工具调用格式转换 | |
| 智能体 | Agent 循环 | agentkits/agent | ReAct 风格智能体 |
| 工具调用 | agentkits/tools | 定义和执行工具 | |
| 多轮对话 | agentkits/conversation | 滑动窗口管理 | |
| 记忆 | agentkits/agent-memory | 短期/长期记忆、自动摘要 | |
| 提示词模板 | agentkits/prompt-template | Handlebars 风格模板 | |
| 安全护栏 | agentkits/guardrails | 输入输出校验、PII 检测 | |
| RAG | RAG 管线 | agentkits/rag | 检索-增强-生成 |
| 重排序 (API) | agentkits/rerank | Cohere / Jina 重排 | |
| 重排序 (本地) | agentkits/reranker | Cross-encoder 本地重排 | |
| 网页搜索 | agentkits/web-search | Brave / Tavily / Serper / SearXNG | |
| PDF 解析 | agentkits/pdf-parser | 文本提取与分块 | |
| 文本分块 | agentkits/chunker | 固定 / 句子 / 递归 / 语义 | |
| 多模态 | 视觉理解 | agentkits/vision | 跨供应商图像理解 |
| 语音合成 | agentkits/tts | OpenAI / Azure / ElevenLabs | |
| 语音识别 | agentkits/stt | 语音转文字 | |
| 图像生成 | agentkits/image | DALL-E / Stability | |
| 基础设施 | 故障转移 | agentkits/failover | 自动供应商切换 |
| 智能路由 | agentkits/router | 按成本/速度/能力路由 | |
| 模型路由 | agentkits/model-router | 规则引擎 | |
| 限流器 | agentkits/ratelimit | Token bucket | |
| 供应商限流 | agentkits/rate-limiter | 按供应商限流 | |
| 响应缓存 | agentkits/response-cache | LRU + TTL | |
| 重试 | agentkits/retry | 指数退避 + 抖动 | |
| 成本计算 | agentkits/cost | 跨供应商定价对比 | |
| Token 计数 | agentkits/token-counter | 按模型精确计数 | |
| 代理服务器 | agentkits/proxy | OpenAI 兼容代理 | |
| 基准测试 | agentkits/benchmark | 延迟与吞吐量 | |
| 供应商测试 | agentkits/test | 一键检测连通性 | |
| 日志 | agentkits/logger | 结构化日志 | |
| 链路追踪 | agentkits/tracing | OpenTelemetry 兼容 | |
| 代码解释器 | agentkits/code-interpreter | 沙箱执行 JS/Python/Shell | |
| 模型评估 | agentkits/evaluation | 多模型输出对比 | |
| 协议 | MCP 客户端 | agentkits/mcp-client | 连接 MCP 服务器 |
| 工作流引擎 | agentkits/workflow | 分支 / 并行 / 重试 | |
| A2A 协议 | agentkits/a2a | Google Agent-to-Agent |
npx agentkits chat [--provider P] [--model M] # 交互式对话
npx agentkits embed "文本" [--provider P] # 生成向量
npx agentkits benchmark [--providers a,b,c] [--runs N] # 供应商对比
npx agentkits test # 测试已配置供应商
npx agentkits cost # 定价对比
npx agentkits serve [--port N] # OpenAI 兼容代理
npx agentkits list # 列出所有供应商
npx agentkits mcp connect <url> # 连接 MCP 服务器
opc-agent 基于 AgentKits 构建，展示了完整的生产级用法：
// opc-agent 内部使用 AgentKits 的模型推荐 + 统一调用
import { recommendModel, createChat, withBrain, checkProvider } from 'agentkits';
// 1. 推荐最适合当前任务的模型
const picks = recommendModel({ task: 'coding', budget: 'low' });
// 2. 创建带记忆的对话
const chat = createChat({ provider: picks[0].provider, model: picks[0].model });
const smartChat = withBrain(chat.complete.bind(chat), { brainUrl: process.env.BRAIN_URL });
// 3. 健康检查确保可用
const health = await checkProvider(picks[0].provider);
if (!health.available) { /* fallback */ }
AgentKits 是跃盟科技 AI 四件套的核心层：
| 层级 | 项目 | 定位 |
|---|---|---|
| 🧠 | DeepBrain | 知识大脑 — 语义存储 + 自动关联 |
| 📊 | AgentKits ← 你在这里 | 模型层 — 统一 API + 推荐 + 记忆增强 |
| 🤖 | OPC Agent | 智能体 — 基于 AgentKits 的生产级 Agent |
| 🔌 | OpenClaw | 运行时 — Agent 编排 + 多端接入 |
DeepBrain (知识) → AgentKits (模型) → OPC Agent (智能体) → OpenClaw (运行时)
# 国际供应商
OPENAI_API_KEY=sk-...
GEMINI_API_KEY=AI...
ANTHROPIC_API_KEY=sk-ant-...
# 中国供应商
DEEPSEEK_API_KEY=sk-... # 深度求索
DASHSCOPE_API_KEY=sk-... # 通义千问
ZHIPU_API_KEY=... # 智谱 AI
MOONSHOT_API_KEY=sk-... # 月之暗面 (Kimi)
YI_API_KEY=... # 零一万物
BAICHUAN_API_KEY=... # 百川智能
SILICONFLOW_API_KEY=... # 硅基流动
STEPFUN_API_KEY=... # 阶跃星辰
MINIMAX_API_KEY=... # MiniMax
Apache-2.0 · Made by Deepleaper 跃盟科技
OpenRouter with Memory: Unified Model API + Smart Recommendation + Memory-Enhanced
One API for all LLMs. Auto-recommend the best model. Every call enhanced with memory.
npm install agentkits
import { recommendModel, createChat, withBrain } from 'agentkits';
// 🎯 Smart recommendation: pick the best model by task + budget
const picks = recommendModel({ task: 'coding', budget: 'medium' });
console.log(picks[0]);
// { model: 'deepseek-coder-v2', provider: 'deepseek', reason: 'Best cost/perf for coding tasks', ... }
// 🤖 Unified API: switch providers by changing one word
const chat = createChat({ provider: 'deepseek', model: 'deepseek-chat' });
const reply = await chat.complete('Write a quicksort in TypeScript');
// 🧠 Memory-enhanced: auto-recall context from DeepBrain
const smartChat = withBrain(chat.complete.bind(chat), { brainUrl: 'http://localhost:3333' });
| Feature | Description | |
|---|---|---|
| 🤖 | Unified Model API | OpenAI / Anthropic / Google / DeepSeek / Qwen / Zhipu / Moonshot / Ollama — one interface |
| 🎯 | Smart Recommendation | recommendModel() picks optimal model by task + budget + speed |
| 💰 | Cost Estimation | estimateModelCost() precise pricing for 20+ models |
| 🏥 | Health Check | checkProvider() real-time availability + latency for 7 providers |
| 🧠 | Memory-Enhanced | withBrain() auto-chains DeepBrain: recall → chat → learn |
| 🎨 | Web UI | Model management dashboard + Playground |
| 📊 | 40 Modules | RAG / Agent / Workflow / MCP / A2A / Vision / TTS / STT / Guardrails |
LLM (19 providers): OpenAI (GPT-4o), Anthropic (Claude 3.5), Google (Gemini 2.5), DeepSeek, DashScope (Qwen), Zhipu (GLM-4), Moonshot (Kimi), Yi, Baichuan, SiliconFlow, StepFun, MiniMax, Grok, Cohere, Fireworks, Together, Groq, Perplexity, Ollama, Custom.
Embedding (15 providers): OpenAI, Google, DashScope, DeepSeek, Zhipu, SiliconFlow, Cohere, Jina, Voyage, Mixedbread, Nomic, Fireworks, Together, Ollama, Custom.
See the Chinese section above for the full pricing table.
// Recommend
recommendModel({ task, budget, speed, local }) → ModelRecommendation[]
estimateModelCost(model, inputTokens, outputTokens) → ModelCostEstimate
// Chat
createChat({ provider, model }) → ChatClient
chat.complete(prompt) → string
chat.stream(prompt) → AsyncIterable<StreamChunk>
// Memory (DeepBrain)
withBrain(chatFn, config) → wrappedChatFn // middleware
recall(query, config) → RecallResult[]
learn(text, config) → void
// Health
checkProvider(provider, config?) → HealthCheckResult
// Web UI
new KitsUI({ port }).start()
// And 40 more modules...
npx agentkits chat | embed | benchmark | test | cost | serve | list | mcp connect
DeepBrain (knowledge) → AgentKits (models) → OPC Agent (agents) → OpenClaw (runtime)
| Layer | Project | Role |
|---|---|---|
| 🧠 | DeepBrain | Knowledge brain — semantic store + auto-linking |
| 📊 | AgentKits ← you are here | Model layer — unified API + recommendation + memory |
| 🤖 | OPC Agent | Agent — production-grade agent built on AgentKits |
| 🔌 | OpenClaw | Runtime — agent orchestration + multi-platform |
Apache-2.0 · Made by Deepleaper 跃盟科技
FAQs
Agent Model Layer — One-line LLM access with built-in memory. 29 providers, 18 embeddings, zero lock-in.
The npm package agentkits receives a total of 48 weekly downloads. As such, agentkits popularity was classified as not popular.
We found that agentkits demonstrated a healthy version release cadence and project activity because the last version was released less than a year ago. It has 1 open source maintainer collaborating on the project.
Did you know?
Socket for GitHub automatically highlights issues in each pull request and monitors the health of all your open source dependencies. Discover the contents of your packages and block harmful activity before you install or update your dependencies.
Security News
npm confirmed a tooling bug incorrectly marked several one-character packages as security holders and said it was working on a rollback.
Research
/Security News
Newer packages in this compromise use native extensions and .pth loaders to execute JavaScript stealers in developer environments.
Research
Socket found 37 malicious PyPI wheels that abuse Python startup hooks to launch a Bun-powered credential stealer tied to Mini Shai-Hulud/Miasma.