Introducing AgentExecutor: A Powerful Framework for AI Agents on UBOS

1. Introduction to OnPrem.LLM’s AgentExecutor

OnPrem.LLM has become a go‑to platform for developers who need large language models (LLMs) that run locally or in the cloud while supporting tool‑calling. The original OnPrem.LLM AgentExecutor documentation showcases a rich set of examples, but the core idea is simple: an agent receives a natural‑language task, decides which tool to invoke, runs the tool, and iterates until the task is marked complete.

For software developers and AI enthusiasts, AgentExecutor offers a unified API that works with any LiteLLM‑compatible model—whether you prefer OpenAI’s GPT‑4, Anthropic’s Claude, Google Gemini, or a self‑hosted Ollama model. This flexibility makes it ideal for building everything from a calculator to a full‑blown financial‑analysis bot.

2. The Nine Default Tools at Your Disposal

When you instantiate AgentExecutor() without specifying enabled_tools, the framework automatically loads nine powerful utilities. These tools are deliberately orthogonal, ensuring a MECE (Mutually Exclusive, Collectively Exhaustive) coverage of common development needs:

• read_file – Load an entire file into memory.
• read_lines – Pull specific line ranges from a file.
• edit_file – Perform find/replace operations.
• write_file – Overwrite or create a new file.
• grep – Search for regex patterns across files.
• find – Locate files using glob patterns.
• run_shell – Execute arbitrary shell commands (optional for security).
• web_search – Query a search engine for up‑to‑date information.
• web_fetch – Retrieve and parse the raw HTML of a URL.

You can disable any of these tools with disable_shell=True or by passing a custom enabled_tools list. This granular control is essential when you need to enforce a strict security posture.

3. Real‑World Usage Scenarios

4. Why Sandboxed Execution Matters

Allowing an LLM to run run_shell can be a double‑edged sword. While it unlocks powerful automation, it also opens a path to unintended file system changes or network calls. AgentExecutor mitigates this risk with two complementary strategies:

1. disable_shell=True – Removes the ability to execute arbitrary commands.
2. sandbox=True – Spins up an isolated Docker or Podman container for the entire agent lifecycle. The container is destroyed automatically, guaranteeing a clean environment for each run.

The sandbox adds a modest 5‑10 seconds of startup overhead, but the security payoff is huge—especially when agents process untrusted prompts or operate on production data.

5. Building and Registering Custom Tools

Custom tools are ordinary Python callables that follow a simple signature (def tool_name(...)) and return JSON‑serializable data. Once defined, they are passed to AgentExecutor(custom_tools=[...]). Below is a concise recipe:

def get_current_stock_price(ticker: str) -> dict:
    import yfinance as yf
    stock = yf.Ticker(ticker)
    info = stock.info
    return {
        "ticker": ticker.upper(),
        "price": round(info.get("currentPrice") or info.get("regularMarketPrice"), 2),
        "timestamp": datetime.utcnow().isoformat()
    }

After registration, the agent can invoke get_current_stock_price just like any built‑in tool. This pattern scales: you can add sentiment analysis, image OCR, or even a proprietary database query engine.

For developers who need a rapid start, UBOS offers a UBOS templates for quick start that include pre‑wired custom‑tool scaffolds, saving hours of boilerplate.

6. How AgentExecutor Fits Into the UBOS AI Platform

UBOS provides a full‑stack UBOS platform overview that unifies model hosting, workflow orchestration, and UI generation. AgentExecutor can be invoked from the Workflow automation studio or the Web app editor on UBOS, turning a single LLM call into a multi‑step business process.

For startups, the UBOS for startups plan includes generous compute credits, making it cheap to prototype AgentExecutor‑driven services. SMBs can leverage the UBOS solutions for SMBs to embed AI agents into CRM, ticketing, or inventory systems without hiring a full data‑science team.

Enterprises looking for scale can adopt the Enterprise AI platform by UBOS, which adds role‑based access control, audit logging, and multi‑region deployment—perfect for regulated industries that still want the agility of autonomous agents.

7. Ready‑Made AgentExecutor Templates You Can Deploy Today

The UBOS Template Marketplace hosts dozens of community‑vetted agents that already integrate custom tools. A few highlights that pair naturally with AgentExecutor workflows:

• AI SEO Analyzer – Scrapes a website, extracts meta data, and suggests optimization steps.
• AI Article Copywriter – Generates long‑form content from a brief, using custom tone‑control tools.
• AI Video Generator – Converts a script into a storyboard, then calls a text‑to‑video engine.

By importing any of these templates into your UBOS workspace, you instantly gain a pre‑configured AgentExecutor pipeline, saving weeks of development time.

8. Cost, Licensing, and Partner Program

UBOS offers transparent UBOS pricing plans that include per‑hour compute, storage, and a generous free tier for hobbyists. For agencies and system integrators, the UBOS partner program provides revenue sharing, co‑marketing, and dedicated technical support.

9. About UBOS

Founded on the principle that AI should be accessible to every developer, About UBOS details a mission to democratize large‑scale model deployment, low‑code orchestration, and AI‑first product design. Their ecosystem includes AI marketing agents, voice synthesis via ElevenLabs AI voice integration, and data indexing with Chroma DB integration.

10. Conclusion – The Power and Flexibility of AgentExecutor

AgentExecutor bridges the gap between raw LLM capabilities and production‑grade automation. By offering a well‑defined set of default tools, sandboxed execution, and a straightforward path to custom extensions, it empowers developers to build trustworthy AI agents that can:

• Generate code, tests, and documentation on demand.
• Perform live web research and synthesize reports.
• Interact with proprietary APIs through custom tools.
• Run safely inside isolated containers, eliminating side‑effects.

When combined with the broader UBOS ecosystem—templates, pricing flexibility, and partner support—AgentExecutor becomes a cornerstone for building next‑generation AI‑driven products. Whether you are a solo developer, a fast‑growing startup, or an enterprise looking to scale autonomous agents, the framework gives you the confidence to iterate quickly while keeping security and cost under control.