> ## Documentation Index
> Fetch the complete documentation index at: https://docs.literalai.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Changelog

# `0.1.0` (November 12th, 2024)

### Improvements

* Added `createPromptVariant` API to create `Prompt` variants
* Modified `createExperiment` to take as input a `promptVariantId` instead of a `promptId`

<Warning>Starting with `0.1.0`, you need to have Literal AI version 0.1.0-beta or above to create experiments.</Warning>

# `0.0.520` (October 28th, 2024)

### Improvements

* Added fields `label`, `status` and `tags` to `getPromptById` API.

<Warning>Starting with `0.0.520`, you need to have Literal AI version 0.0.629-beta or above.</Warning>

# `0.0.519` (October 21st, 2024)

### Improvements

* Added `release` parameter to Literal AI client constructors

# `0.0.518` (September 18th, 2024)

### Bug fixes

* Use getters to lazy load modules

# `0.0.517` (September 16th, 2024)

### Improvements

* Added link to prompt as URL in the `get_prompt` API
* Introduced decorators with `metadata`, `tags` and `stepid` fields
* Adjusted integrations to account for `metadata` and `tags` logging

### Bug fixes

* Made third-party dependencies optional when importing Literal AI client.

# `0.0.515`

### Improvements

This version improves compatibility with the LangChain ecosystem, notably with complex LangGraph flows.

# `0.0.514`

### Improvements

Simplify the Vercel AI SDK integration by leveraging the internal context.
`threadId` and `parentId` are now automatically handled without the need to pass them as a `literalAiParent` argument.

# `0.0.513`

### Improvements

* Added option to send a root run ID upon creation of a `Step`
* The API `update_prompt_ab_testing` replaces the old `promote_prompt` and extends on it to allow for A/B testing

<Warning>Starting with `0.0.513`, you need to have Literal AI version `0.0.617-beta` or above.</Warning>

# `0.0.512`

### Breaking Changes

Instantiation of the TypeScript client was altered to take in an object as argument:

<CodeGroup>
  ```typescript >= 0.0.512 theme={null}
  client = new LiteralClient({
    apiKey: process.env.LITERAL_API_KEY,
    apiUrl: process.env.LITERAL_API_URL
  });
  ```

  ```typescript < 0.0.512 theme={null}
  client = new LiteralClient(process.env.LITERAL_API_KEY, process.env.LITERAL_API_URL);
  ```
</CodeGroup>

### New Features

This release introduces a new, wrapper-aware method to create attachments : `client.api.createAttachment`. This method is a wrapper around the `uploadFile` method and the `new Attachment` constructor. It allows you to create an attachment in a single call, and attach it to the thread/step in the current context.

The old method using `uploadFile` and `new Attachment` is considered obsolete, although it remains in the SDK for backward compatibility.

### Improvements

* `threadId` is now optional when creating an attachment or uploading a file
* `createAttachment` and `uploadFile` now accept file data in the following formats : ReadableStream, ReadStream, Buffer, File, Blob & ArrayBuffer
* `Generation` creation now supports passing `tags` and `metadata`
* `Experiment` become optionally linked to a `Dataset`

<Warning>Starting with `0.0.512`, you need to have Literal AI version 0.0.615-beta or above.</Warning>

# `0.0.510`

### Breaking Changes

Revamp of the OpenAI instrumentation : instead of manually instrumenting each call result, you can now instrument OpenAI globally. Each subsequently call will be logged with no additional code.

This new implementation is context-aware and will log the OpenAI calls in the right steps and threads if you use the new wrapper syntax.

<Warning>
  If you were using the `instrumentation.openai` method in your code and upgrade the Literal AI SDK to `0.0.510`, you will get the following error at compile time : `error TS2554: Expected 0-1 arguments, but got 2.`.

  To fix this, you should move the `instrumentation.openai` call to the top of your code and remove the arguments apart from the optional `{ tags: []}` argument.

  You can find code examples in the [OpenAI Instrumentation](/integrations/openai) guide.
</Warning>

# `0.0.509`

### New Features

* New syntax with wrappers for `step` and `thread` methods.

# `0.0.504`

### Deprecations

* `format` is deprecated. `formatMessages` should be used now.

### New Features

* Add support for tags with OpenAI Instrumentation
* Allow thread upsert with one single object argument

# `0.0.503`

### Deprecations

* `createPrompt()` is deprecated. `getOrCreatePrompt()` should now be used.

### New Features

* `getOrCreatePrompt()`. For creating a new Prompt, use `getOrCreatePrompt()`.

A Prompt is fully defined by its name, template\_messages, settings and tools. If a prompt already exists for the given arguments, it is returned. Otherwise, a new prompt is created.

```jsdoc theme={null}
/* 
* @param name The name of the prompt to retrieve or create.
* @param templateMessages A list of template messages for the prompt.
* @param settings Optional settings for the prompt.
* @returns The prompt that was retrieved or created.
*/
```