0.6.0

What's Changed

• fix: correct stream completion detection in OpenAI response by @gbfansheng in #19
• Add Anthropic API support (Claude) for code completion by @bastianX6 in #21

New Contributors

• @gbfansheng made their first contribution in #19
• @bastianX6 made their first contribution in #21

Full Changelog: 0.5.0...0.6.0

0.5.0

• Support Ollama FIM API
• Support setting the prompt as raw when the strategy is set to Fill-in-the-middle
• Update the Mistral FIM API to be more inclusive so that it can also support APIs like the DeepSeek one.
• Add error toast in app.

0.4.0

• Fill-in-the-middle strategy now supports custom prompt template so you can use it with other models.
• Support models with an FIM API, for example, Mistral AI FIM endpoint.
• Support setting suggestion token limit.
• OpenAI (compatible) APIs will now use stream response. So it will response more promptly when you give it a line limit.

0.3.0

• Support setting a line limit to suggestions.
  Running LLMs locally can be slow, if you are having this issue, you can, for example, set it to 1 to get a faster full line suggestion.

0.2.0

• Supports Ollama when setting up custom models.
• Added CodeLlama Fill-in-the-Middle strategy, which works very well. It's recommended to be used along with codellama:xb-code series models. You can try other models that support FIM, too. But I am not sure if they share the same prompt format.
• Support trimming repeated suffixes from the response #2
• When Tabby is used, the app will use its own strategy.
• Fix response parsing in strategies.

0.1.0

Initial release!