Migration

Migrating to v1.0.x

All Constructors Require Explicit base_url

In v1.0.0 the API was redesigned so every provider constructor takes an explicit base_url. This removes hidden defaults and makes multi-tenant / proxy routing obvious at the call site.

// Before (v0.x)
let client = LlmClient::openai("sk-...")?;
let client = LlmClient::ollama()?;
let client = LlmClient::openai_with_base_url("sk-...", "https://...")?;

// After (v1.x)
let client = LlmClient::openai("sk-...", "https://api.openai.com/v1")?;
let client = LlmClient::ollama("http://localhost:11434")?;
// openai_with_base_url is removed — just use openai(key, url)

All affected constructors:

Provider	Old	New
openai	openai(key)	openai(key, base_url)
anthropic	anthropic(key)	anthropic(key, base_url)
aliyun	aliyun(key)	aliyun(key, base_url)
zhipu	zhipu(key)	zhipu(key, base_url)
ollama	ollama()	ollama(base_url)
deepseek	deepseek(key)	deepseek(key, base_url)
moonshot	moonshot(key)	moonshot(key, base_url)
volcengine	volcengine(key)	volcengine(key, base_url)
xiaomi	xiaomi(key)	xiaomi(key, base_url)
google	google(key)	google(key, base_url)
tencent	tencent(id, key)	tencent(id, key, base_url)

Deserialize on Protocol Request Types (v1.0.3)

OpenAIRequest, AnthropicRequest, GoogleRequest, OllamaChatRequest and all their nested structs now derive Deserialize in addition to Serialize. This enables reverse proxy / middleware / observability use cases where incoming wire-format bodies need to be parsed.

No breaking change — purely additive.

---

Migrating to v0.5.0

Multi-Modal Content

v0.5.0 introduced native multi-modal support, changing the Message.content field.

Breaking Change:

• Message.content is now Vec<MessageBlock> internally.
• Use helper constructors instead of direct field access.

// Before
let msg = Message { role: Role::User, content: "hello".to_string(), .. };

// After
let msg = Message::user("hello");
// or
let msg = Message::text(Role::User, "hello");