Interested in using optillm as a toolset system for agents

[surfingkaka]

Hi, Thank you for the library, looks nice. I am researching toolsets that agents can use, and the plugins look nice but I was going through the code, for example, the execute and memory plugins, they havent been worked on in a year. Are they feature complete ? Is there a roadmap for maturity ? thank you

[codelion]

Hi @surfingkaka Thanks for your interest in optiLLM!

Great question about the plugins. Let me address your concerns:

The executecode and memory plugins you mentioned are feature complete for their intended scope - they provide basic implementations of code execution and short-term memory capabilities:

• executecode_plugin.py: Provides basic Python code execution in a sandboxed environment
• memory_plugin.py: Implements simple short-term memory for maintaining context across conversations

These plugins were designed as reference implementations to demonstrate the plugin architecture and provide baseline functionality. They work well for their intended use cases and haven't needed major updates because they accomplish what they were designed to do.

However, if you're looking to build a robust agent toolset system, optiLLM offers much more flexibility:

1. MCP (Model Context Protocol) Support

optiLLM has built-in support for https://modelcontextprotocol.io/, which gives you access to a rich ecosystem of tools:

• Configure MCP servers in ~/.optillm/mcp_config.json
• Connect to any existing MCP server (filesystem, databases, APIs, etc.)
• Use community-built MCP tools without writing any code
• The mcp_plugin.py handles all the integration

Example MCP configuration:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/allowed/files"]
    },
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {"GITHUB_TOKEN": "your-token"}
    }
  }
}

2. Build Your Own Plugins Easily

The plugin system is designed to be extremely simple - you can create custom plugins in just a few lines of Python:

Minimal plugin template:

# optillm/plugins/my_custom_plugin.py
SLUG = "mycustom"

def run(system_prompt: str, initial_query: str, client, model: str, 
        request_config: dict = None):
    # Your custom logic here
    # Access databases, APIs, external tools, etc.

    response = client.chat.completions.create(
        model=model,
        messages=[
            {"role": "system", "content": system_prompt},
            {"role": "user", "content": initial_query}
        ]
    )

    return response.choices[0].message.content, response.usage.completion_tokens

That's it! Your plugin is now accessible via model="mycustom-gpt-5-mini" or extra_body={"optillm_approach": "mycustom"}.

3. Existing Advanced Plugins

OptiLLM already includes several sophisticated plugins you can use as references or building blocks:

• deep_research: Web search and research capabilities
• spl: System Prompt Learning for evolving strategies
• genselect: Quality-based response selection
• longcepo: Long-context processing with divide-and-conquer

While we don't have a formal public roadmap, the project is actively maintained:

• Core inference optimization approaches are stable and well-tested
• Plugin architecture is mature and extensible
• Active community contributions and bug fixes
• Recent additions like MARS, deep research, and improved MCP support

The philosophy is: provide solid building blocks and let users compose them as needed rather than trying to be everything to everyone.