What this solves

Function calling turns an LLM into an executor: structured args, validated actions and logged results. RemoteAgent Online is LLM-agnostic: one policy, one audit trail, any model underneath. This guide covers what works in practice, with the safety, transparency and cost controls built in.

Safe, transparent, cheap - engineered in

  • Safe: least-privilege capability grants and local policy enforcement - the device says no before a bad action exists.
  • Transparent: every action, token and tool call lands in a hash-chained, tamper-evident audit log you can verify.
  • Cheap: any LLM - GPT, Claude, Gemini, DeepSeek, Llama, Mistral, Qwen, local models. Your keys, hard caps, no markup.

How RemoteAgent Online solves it

Tool calls are the backbone of the audit log.

  1. Install the open-source agent runtime on your machine or server.
  2. Write the policy - what the agent may touch and what it must never do.
  3. Grant capabilities narrowly: files, tools, APIs, budgets, time windows.
  4. Run the task and audit every step in the log.

Runs on any LLM: GPT-4o, Claude, Gemini, DeepSeek R1, Llama 3, Mistral, Qwen, Phi-3, Ollama local models.

FAQ

Which models support function calling?

Is it safe?

Yes - policies are enforced on the device itself, transport is mTLS-encrypted, and the audit log is hash-chained so nobody can silently rewrite history.

Is it cheap?

Yes - you bring your own LLM keys or run open models locally. No per-seat tax, no token markup, hard budget caps.

Go deeper