v0.76.0

PromptingTools v0.76.0

Diff since v0.75.0

Breaking Changes:

• claude alias now points to the latest Claude 4 Sonnet

Added

• Added support for Anthropic's new web search tool and code execution tool (alias :web_search and :code_execution, respectively). See ?aitools for more information.

Updated

• Updated claude alias to point to the latest Claude 4.0 Sonnet model (same as claudes).

Commits

Merged pull requests:

• Fix Vitepress nav config (#293) (@svilupp)
• Change model for claude alias + add more pre-built tools (#294) (@svilupp)

v0.75.0

PromptingTools v0.75.0

Diff since v0.74.3

BREAKING CHANGES:

• Removed the RAGTools module. Please use the standalone RAGTools.jl package. See the migration guide in the docs & FAQ.
• Default chat model changed to GPT 4.1 mini (previously it was GPT 4o mini, variable PromptingTools.MODEL_CHAT).

Added

• Added Claude 4 models (claude-sonnet-4-20250514, claude-opus-4-20250514) and set them as new defaults for aliases claudeo and claudes (Added specs for the new Claude 4 text editor as a ToolRef :str_replace_based_edit_tool).
• Added a Deepwiki.com badge for AI-powered repo documentation.
• Added Gemini 2.5 Flash Preview model from 20th May 2025 (alias gem25f).

Updated

• Updated the OllamaManagedSchema to accept URLs with and without protocol to allow endpoints with https:// (previously only http:// was supported).
• Removed the RAGTools module.

Fixed

• Fixed @ai_str macro to keep a dynamic reference to your default model (PromptingTools.MODEL_CHAT). Prevents "baking in" the default model if PromptingTools is precompiled by another package.

### Commits
Many first-time contributors - thank you!

Merged pull requests:

• Add example of using Responses API (#286) (@svilupp)
• Remove Experimental.RAGTools (#288) (@svilupp)
• Fix default model insertion in ai_str macros (#290) (@omlins)
• Add Claude 4 models (#291) (@svilupp)
• Remove hard-coded http for ollama (#292) (@jochalek)

Closed issues:

• Change default model at runtime in precompiled package (#289)

v0.74.3

PromptingTools v0.74.3

Diff since v0.74.2

No breaking changes.

Added

• Added support for Gemini 2.5 Pro and Flash preview models (aliases gem25p and gem25f, respectively).

Commits

Merged pull requests:

• Added Gemini 2.5 Pro/Flash models (#285) (@svilupp)

v0.74.2

PromptingTools v0.74.2

Diff since v0.74.1

No breaking changes.

Added

• Added support for OpenAI's O3 and O4-Mini reasoning models (aliases o3 and o4m, respectively).

Fixed

• Fixed a bug in the precompilation. If a user set MODEL_CHAT to be a non-OpenAI model, aiclassify would throw an error as it could not validate the tokenizer for correct functionality. Bypassed by forcing a specific tokenizer vocabulary (irrelevant for precompilation purposes).
• Fixed failing CI due to not publishing docs for CustomRetryLayer.

Commits

Merged pull requests:

• Fix precompilation bug (#282) (@svilupp)
• Add OpenAI o3 and o4 models (#283) (@svilupp)

Closed issues:

• Precompile PromptingTools.jl Failed after executing the command 'PT.set_preferences!("PROMPT_SCHEMA" => "OllamaSchema", "MODEL_CHAT"=>"llama2")' (#281)

v0.74.1

PromptingTools v0.74.1

Diff since v0.74.0

No breaking changes.

Added

• Added support for GPT-4.1 models (aliases gpt41 for GPT 4.1, gpt41m for GPT 4.1 mini, gpt41n for GPT 4.1 nano; see details here).

Commits

Merged pull requests:

• Add GPT 4.1 models (#280) (@svilupp)

v0.74.0

PromptingTools v0.74.0

Diff since v0.73.0

Breaking

Added

• Added support for extended output (up to 128K tokens) with Claude 3.7 models via the :extended_output beta header.

Fixed

• Added validation to ensure that thinking.budget_tokens does not exceed max_tokens when using Anthropic API.

Commits

Merged pull requests:

• Adding in the official sonnet model name (#276) (@Sixzero)
• Add extended output support for Claude 3.7 models (#278) (@Sixzero)

v0.73.0

PromptingTools v0.73.0

Diff since v0.72.3

Breaking changes

Added

• Added a new enable_retry! function to enable the retry layer for HTTP requests. Main use case is to prevent rate limiting errors from external AI services, but you can fully customize the retry behavior for any HTTP status codes, etc. See ?enable_retry! for more information.

Updated

• Updated the StreamCallbacks dependency to 0.6 to improve parsing of streaming responses with thinking from Anthropic.

Commits

Merged pull requests:

• Add custom retry layer (#273) (@svilupp)
• Update version (#275) (@svilupp)

v0.72.3

PromptingTools v0.72.3

Diff since v0.72.2

Breaking Changes

Added

• Added OpenAI's GPT 4.5 research preview (alias gpt45) to the model registry.

Commits

Merged pull requests:

• Add OpenAI's GPT 4.5 (#274) (@svilupp)

v0.72.2

PromptingTools v0.72.2

Diff since v0.72.1

Breaking Changes

Added

• Added Claude 3.7 Sonnet model to the model registry (alias claude37 and updated default claude alias to point to it).

Commits

Merged pull requests:

• Add Claude 3.7 model support (#271) (@Sixzero)

v0.72.1

PromptingTools v0.72.1

Diff since v0.72.0

Added

• Added Qwen models from OpenRouter API: Qwen Turbo, Plus and Max models with competitive pricing and 128K context window (aliases orqwenturbo, orqwenplus and orqwenmax, respectively).

Commits

Merged pull requests:

• feat: add Qwen models from OpenRouter API (#269) (@Sixzero)
• Update alias for open router models for consistency (#270) (@svilupp)