Feature Request: Telemetry Controls and Local (llama.cpp) Support

Summary

Two areas need clarification and improvement:

1. Telemetry behavior and controls
2. Local model support (llama.cpp and compatible backends)

1. Telemetry

Problem

There is no clear documentation describing:

• Whether telemetry exists
• What data is sent
• How to disable it

Request

• Document:
  • What data (if any) is collected
  • When it is sent
  • Destination endpoints
• Provide a full opt-out, e.g.:
  • DISABLE_TELEMETRY=1
• Document all related environment variables
• Add a mode where no outbound requests are made except the configured model endpoint

Optional

• CLI flag: --no-telemetry
• Debug option to log outbound requests

2. Local Model Support (llama.cpp)

Problem

Local usage via OpenAI-compatible endpoints is not documented and may not perform well.

Request

• Document setup for:
  • llama.cpp server
  • LM Studio
  • other OpenAI-compatible backends
• Provide a working example configuration

Performance / Control

• Add a low-overhead mode that:
  • Reduces system prompt size
  • Limits tool usage
  • Avoids unnecessary context changes
• Allow:
  • Disabling tools
  • Controlling prompt layers
  • Limiting agent loop iterations

Compatibility

• Ensure support for:
  • Streaming responses
  • Tool/function calling formats used by local backends
• Provide fallback when tool calling is unavailable

No files, tests, or entry points are named. Start by locating the telemetry, environment-variable, CLI configuration, and OpenAI-compatible model request paths; then determine the scope of documentation, opt-out controls, local backend compatibility, and agent-loop changes. Done requires agreed behavior, documented setup, and verified streaming and tool-calling fallbacks.