Ruboty::AiAgent

A Ruboty handler that uses LLMs (e.g., OpenAI) and MCP (Model Context Protocol) tools to generate AI-assisted replies. When an incoming message does not match any other handler, this plugin responds via an LLM and can call MCP tools when available.

Add to your Gemfile:

gem 'ruboty'
gem 'ruboty-ai_agent'

Then install:

bundle install

Configure environment variables (e.g., via .env):

• OPENAI_API_KEY: Your LLM provider API key (required)
• OPENAI_MODEL: Model name (optional)

• Any message that does not match other handlers is sent to the AI and replied to.
• Conversations are tracked per thread, and tools exposed via MCP servers can be invoked through function calls when the model requests them.

Conversation management:

• /clear — Clear the current thread’s conversation history
• /compact — Summarize and compact the conversation history

System prompt (work in progress):

• set system prompt "<PROMPT>" — Set the system prompt (WIP)
• show system prompt — Show the current system prompt (WIP)

AI memory (lightweight profile):

• add ai memory "<TEXT>" — Add a memory entry
• remove ai memory <INDEX> — Remove a memory by index
• list ai memories — List all memory entries

AI custom commands (work in progress):

• add ai command /<NAME> "<PROMPT>" — Add a custom AI command (WIP)
• remove ai command /<NAME> — Remove a custom AI command (WIP)
• list ai commands — List available built-in commands

MCP (Model Context Protocol):

• add mcp <NAME> <OPTIONS> <URL> — Add an MCP server
  • Example (HTTP transport with auth header):
    • add mcp search --transport http --header 'Authorization: Bearer xxx' https://example.com/mcp
    • add mcp search --transport http --bearer-token xxx https://example.com/mcp
  • Options:
    • --transport http|sse (currently only http implemented; sse is not yet implemented)
    • --header 'Key: Value' (repeatable)
    • --bearer-token <TOKEN> (shorthand for --header 'Authorization: Bearer <TOKEN>')
• remove mcp <NAME> — Remove an MCP server
• list mcp / list mcps — List configured MCP servers

• Unmatched messages are handled by Ruboty::Handlers::AiAgent#chat and sent to the configured LLM.
• If MCP servers are configured, their tool definitions are exposed to the model; when the model requests a function call, the corresponding MCP tool is invoked and its response is fed back into the conversation.
• During tool calls, the bot streams short status updates and returns a brief summary of the tool’s response.
• Conversation data is stored in Ruboty’s brain under the :ai_agent namespace.

Install dependencies:

bin/setup

Local run with dotenv (for development):

bundle exec ruboty --dotenv

# or using the included launcher
bin/ruboty --dotenv

Run tests:

rake spec

Interactive console:

bin/console

Install locally:

bundle exec rake install

Release: Update version.rb and run bundle exec rake release to tag, push, and publish to RubyGems.

Bug reports and pull requests are welcome on GitHub at https://github.com/tomoasleep/ruboty-ai_agent.

Everyone interacting in the Ruboty::AiAgent project’s codebases, issue trackers, chat rooms and mailing lists is expected to follow the code of conduct.