Not a replacement story
不是替代关系

QVeris vs LiteLLM: Gateway or Capability Layer?QVeris vs LiteLLM：模型网关还是能力层？

The useful comparison is not “which one wins.” LiteLLM belongs in the model layer. QVeris belongs in the capability layer. Production agents usually need both: one to reach models, one to reach the real world.

这个对比的重点不是“谁赢”。LiteLLM 属于模型层，QVeris 属于能力层。生产级 Agent 通常两者都需要：一个连接模型，一个连接真实世界。

See the architecture查看架构Choose the right layer选择正确层

If LiteLLM answers “which model should respond?”, QVeris answers “which verified capability should the agent call?”

如果 LiteLLM 回答的是 “该调用哪个模型？”，QVeris 回答的是 “Agent 该调用哪个已验证能力？”

Why This Comparison Gets Misread

为什么这个对比容易被误读

Most “vs” searches assume two products are competing for the same slot. That is not the right mental model here. LiteLLM is infrastructure for model traffic. It helps teams normalize providers, control keys, apply budgets, and route requests across OpenAI-compatible model endpoints.

大多数 “vs” 搜索都默认两个产品在争同一个位置。但这里不能这样理解。LiteLLM 是模型流量基础设施，帮助团队统一供应商、控制 Key、设置预算，并在 OpenAI 兼容模型端点之间路由请求。

QVeris starts after a different question. Once an agent has a goal, it often needs current market data, a filing, an external calculation, a document parser, a search tool, or another real-world capability. A model gateway can help choose the model, but it does not automatically give the model reliable tools or verified data. QVeris is designed for that second layer.

QVeris 从另一个问题开始：当 Agent 有了目标，它往往需要当前市场数据、文件、外部计算、文档解析、搜索工具或其他真实世界能力。模型网关能帮你选择模型，但不会自动给模型可靠工具和已验证数据。QVeris 解决的是第二层。

Two Layers, Two Jobs

两层架构，两种职责

A model gateway and a capability routing layer can sit side by side. They do not compete for the same job.

模型网关和能力路由层可以并排存在。它们不是在争同一件事。

Model layer
Agent promptQuestion, task, context, model preference.
LiteLLMProvider routing, budgets, keys, fallback.
LLM responseReasoning, generation, summarization.
Capability layer
Agent intentNeed live data, tool action, or external service.
QVerisDiscover, inspect, call verified capabilities.
Structured resultJSON from data, tools, documents, or services.

Where Each Product Fits

每个产品适合放在哪里

The easiest way to choose is to name the bottleneck. If the bottleneck is model access, use the model layer. If the bottleneck is trustworthy real-world action, use the capability layer.

最简单的选择方式是先说清瓶颈。如果瓶颈是模型接入，就用模型层；如果瓶颈是可信的真实世界动作，就用能力层。

You are standardizing LLM providers你在统一 LLM 供应商LiteLLM is the better fit. It gives your team a proxy layer for model requests, provider routing, fallback, virtual keys, and budget controls.LiteLLM 更适合。它为模型请求提供代理层，处理供应商路由、fallback、虚拟 Key 和预算控制。
Your agent needs live information你的 Agent 需要实时信息QVeris is the better fit. It lets the agent discover and inspect capabilities before calling live data, tools, and external services.QVeris 更适合。它让 Agent 在调用实时数据、工具和外部服务之前，先发现并检查能力。
You are moving from demo to production你正在从 demo 走向生产Use both. LiteLLM keeps model access controlled; QVeris gives the agent auditable capability execution and structured outputs.两者都用。LiteLLM 保持模型接入可控，QVeris 提供可审计的能力执行和结构化输出。

Use LiteLLM when the problem is model access

问题是模型接入时，用 LiteLLM

LiteLLM is the right layer when your team needs an OpenAI-compatible proxy, self-hosting, provider config, virtual keys, budgets, and model fallback.

当团队需要 OpenAI 兼容代理、自托管、供应商配置、虚拟 Key、预算和模型 fallback 时，LiteLLM 是正确层。

Use QVeris when the problem is real-world capability

问题是真实世界能力时，用 QVeris

QVeris is the right layer when agents need verified live data, tools, documents, market intelligence, or external actions with structured outputs.

当 Agent 需要已验证实时数据、工具、文档、市场情报或外部动作，并且需要结构化输出时，QVeris 是正确层。

A Production Agent Workflow

一个生产级 Agent 工作流

A realistic agent does not only produce text. It reasons, asks for data, checks schemas, calls tools, and returns a structured result. That is where the two layers become complementary.

真实的 Agent 不只是生成文本。它会推理、请求数据、检查 schema、调用工具，并返回结构化结果。这时两层会自然互补。

01Reason with a model先用模型推理The agent sends a prompt through LiteLLM to the selected model.Agent 通过 LiteLLM 把 prompt 发送到选定模型。
02Identify missing context识别缺失上下文The model decides it needs live data, a filing, a parser, or another capability.模型判断它需要实时数据、文件、解析器或其他能力。
03Inspect before calling调用前先检查QVeris returns capability schemas, quality signals, cost, and provider context.QVeris 返回能力 schema、质量信号、成本和供应商上下文。
04Return structured output返回结构化输出The capability call returns JSON that the agent can cite, transform, or pass back to the model.能力调用返回 JSON，Agent 可以引用、转换或传回模型。

Decision Map

决策地图

If your agent needs...Better layerWhy
GPT, Claude, Gemini, or Llama routingLiteLLMThat is model-provider normalization.
Self-hosted model traffic policyLiteLLMThe gateway runs under your infrastructure controls.
Live data, verified tools, or document actionsQVerisThat is a capability call, not a model call.
Production agent workflowsBothModels reason; capabilities give them trustworthy context and actions.

Frequently Asked Questions

常见问题

Is QVeris a LiteLLM alternative?QVeris 是 LiteLLM 替代品吗？

Not directly. LiteLLM is a model gateway. QVeris is a capability routing layer.

不是直接替代。LiteLLM 是模型网关，QVeris 是能力路由层。

Can they work together?它们能一起用吗？

Yes. LiteLLM routes model calls, while QVeris routes calls to tools, data, and services.

可以。LiteLLM 路由模型调用，QVeris 路由工具、数据和服务调用。

Do I still need a model gateway?还需要模型网关吗？

Usually yes, if you care about provider control, fallback, budgets, or model routing.

通常需要，尤其当你关心供应商控制、fallback、预算或模型路由时。

When should I add QVeris?什么时候加 QVeris？

Add it when your agent needs current data, verified execution, or external capabilities.

当 Agent 需要当前数据、已验证执行或外部能力时，就该加 QVeris。

What is the difference between model routing and capability routing?模型路由和能力路由有什么区别？

Model routing chooses an LLM provider. Capability routing chooses the tool, data source, or service an agent should call.

模型路由选择 LLM 供应商；能力路由选择 Agent 应该调用的工具、数据源或服务。

Does QVeris return model responses?QVeris 返回模型回复吗？

QVeris focuses on capability execution and structured results. It is designed to work beside your model layer.

QVeris 关注能力执行和结构化结果，设计上是和模型层并排工作。