chatpharo

Agents (model backends)

ChatPharo routes requests to a selected “agent”. All agents implement the ChatPharoAgent contract so the UI and chat/session code stay backend-agnostic.

Core responsibilities

An agent typically maintains:

• model
• systemPrompt
• history (LLM-friendly log)
• optional tool calling support
• optional thinking content support

Included agents (overview)

Agent	Category	Notes
ChatPharoNullAgent	Offline	Returns a message telling you to configure a backend
ChatPharoOllamaAgent	Local	Talks to an Ollama server (localhost:11434 by default)
ChatPharoLMStudioAgent	Local	Talks to a local LM Studio server
ChatPharoClaudeAgent	Cloud	Tool-call loop with max iterations; thinking content supported
ChatPharoGeminiAgent	Cloud	General-purpose cloud backend
ChatPharoDeepSeekAgent	Cloud	Code-focused cloud backend
ChatPharoMistralAgent	Cloud	Present but marked as not supported yet

Tool calling loop

Some agents follow this pattern:

1. Send request to backend.
2. If the response includes tool_calls, execute tools and append results to history.
3. Call the backend again.
4. Stop when no tool calls remain, or when maximumIterations is reached.

Related docs

• Tool schema + execution: tools.md
• External API warnings: safety-ethics.md

This site is open source. Improve this page.