Skip to content

Conversation

@zavocc
Copy link
Owner

@zavocc zavocc commented Aug 15, 2025

Key highlights

  • New models:
    gemini-2.5-flash, gemini-2.5-pro, gemini-2.5-flash-nonthinking, grok-3-latest, kimi-k2-instruct from groq, gpt-5 and gpt-5-mini including it's high variant, o4-mini and it's high variant, Deepseek-R1, Deepseek-V3, claude-sonnet-4

  • Removed models:
    gpt-4o, gpt-4.1, gpt-4.1-mini, o3-mini, o1, LLaMA DeepSeek R1, Qwen 32B DeepSeek R1

  • Updated system instructions to be more saas

  • Tool use support for OpenRouter, Anthropic, OpenAI, and Kimi models

  • Allow agentic loop behavior when using tools in all models

  • New and merging tools:

    • Audio Tools: podcastgen, audio_generator_gemini, audio_generator
    • Web Search: web_search
    • Ideation Tools: canvas, artifacts
    • ImageGen: image_generator
    • GitHub: github_file_tool, github_search_tool
    • YouTube: youtube_search, youtube_corpus

Organized tools for it to automatically register in tools/ToolName/tool.py if there's a schema and functions

  • Removed tools:

    • BingSearch: bing_search, url_extractor
  • Refactorization of codebase such as including new helper abstract base classes and provider plugging

  • New modalities:

    • Claude models: PDF inputs
    • OpenRouter: PDF and image inputs
  • Cut off support for music features for now

zavocc and others added 30 commits April 18, 2025 06:39
- Now it's organized into its own dedictaed module directory wih manifest and tool implementation, and get_tools_list() method will perform checks if it's valid
- New naming convention
- Deprecate tools.yaml file and the get_tools_list() generator now searches for valid tools found and populates them
TODO: handle agentic tool use part both Gemini and Claude
Since it will error as its not properly handled
… and had redundant logic otherwise)

Code Execution as a tool
…erminal which degrades performance with binary data
- Thinking mode (auto budget set)
- Non-thinking mode

Update to Grok 3
https://learn.microsoft.com/en-us/azure/ai-foundry/model-inference/how-to/create-model-deployments?pivots=ai-foundry-portal

Allows model switching with a single endpoint
currently Azure AI foundry can work with deployments

- Azure openai
- Managed compute
- Serverless compute
but it is tied to specific models
while the new inference service endpoint is preview, it would enable us to easily use multiple models in one endpoint
Update Bing tool
- Seems the new one flags almost all queries
TODO: check sysprompt
@zavocc zavocc merged commit b9a778e into master Aug 15, 2025
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants