OnPrem.LLM: Launch Autonomous AI Agents with Sandboxed Execution in 2 Lines of Code

OnPrem.LLM has introduced the AgentExecutor — a pipeline for creating autonomous AI agents that can execute complex tasks using a variety of tools, with support for both cloud and local models.

How It Works

The AgentExecutor is implemented using their coding agent, PatchPal, and provides a simple interface for launching AI agents with sandboxed execution:

from onprem.pipelines import AgentExecutor

# Full access (all tools including shell):
executor = AgentExecutor(model='anthropic/claude-sonnet-4-5')

# Safer mode (no shell access):
executor = AgentExecutor(model='openai/gpt-5-mini', disable_shell=True)

Supported Models

The pipeline works with any LiteLLM-supported model that supports tool-calling:

• Cloud: openai/gpt-5.2-codex, anthropic/claude-sonnet-4-5, gemini/gemini-1.5-pro
• Local: Ollama (ollama/llama3.1), vLLM (hosted_vllm/), llama.cpp (via OpenAI interface)

9 Built-in Tools

1. read_file — Read complete file contents
2. read_lines — Read specific line ranges from files
3. edit_file — Edit files via find/replace
4. write_file — Write complete file contents
5. grep — Search for patterns in files
6. find — Find files by glob pattern
7. run_shell — Execute shell commands
8. web_search — Search the web for information
9. web_fetch — Fetch and read content from URLs

Security Features

• Working directory isolation — agents cannot read or write outside their specified working directory
• Shell access is optional — can be disabled for safer deployments
• Tool whitelisting — users can specify exactly which tools an agent can access
• Sandboxed execution — agents operate within defined boundaries

Use Cases

The system is designed for tasks like autonomous code generation, file manipulation, web research, and complex multi-step workflows — all from a local environment without sending data to external services (when using local models).

Source: OnPrem.LLM Documentation | GitHub | HN: 48pts