Toolofficial publisher

Tools OpenAI Agents SDK

Tool : A precise action the AI can call · requires openai

WebSearchTool, FileSearchTool, ComputerTool and your own functions.

OpenAI's agents SDK with its included hosted tools.

Docs

Overall score

6.9 /10

Security: one caveat, see the card

Utility

8.0

Reliability

6.0

Security

One caveat

Ease

5 min · Technical

Security reflects where data goes, requested permissions and incidents on record at the verification date. It is not a guarantee of zero risk: check the deployment mode and the connections you enable.

What it does, concretely

Building an agent that can act on the world, not just answer questions, requires giving it access to tools: web search, file retrieval, code execution, external API calls. OpenAI's Agents SDK centralises these tools in a single coherent framework, available in Python and TypeScript.

The SDK distinguishes five categories: OpenAI-hosted tools (WebSearchTool, FileSearchTool, CodeInterpreterTool, HostedMCPTool, ImageGenerationTool, ToolSearchTool, ProgrammaticToolCallingTool), local or container tools (ComputerTool, ShellTool, ApplyPatchTool), Python functions wrapped as tools (FunctionTool), agents called as tools, and the experimental Codex tool. Every category integrates into the same Runner lifecycle, with guardrails, sessions, and approval handling.

This SDK targets developers building autonomous or semi-autonomous agents who want to orchestrate multiple tool types without rewriting the plumbing each time. It suits both simple workflows and complex pipelines combining search, image generation, and code execution.

It is not suited to someone looking for a graphical interface or a ready-made product: everything is configured in code, and API costs accumulate depending on which tools are activated.

What it offers

·WebSearchTool lets the agent search the web, with options for filters, user location, context size, and image results.
·FileSearchTool queries your OpenAI Vector Stores with up to 50 results, configurable filters, and ranking options.
·CodeInterpreterTool lets the model execute code in an OpenAI-hosted sandbox environment.
·HostedMCPTool exposes a remote MCP server's tools directly to the model.
·ImageGenerationTool generates images from a prompt without leaving the agent's flow.
·ToolSearchTool lets the model load only the tools it needs at runtime from a large surface, reducing tool-schema token consumption.
·ProgrammaticToolCallingTool lets the model generate JavaScript to coordinate multiple tool calls in parallel or in loops inside a hosted V8 environment.
·ShellTool supports shell command execution locally or in an OpenAI-hosted container, with support for reusable mounted skills.

Our tips

For WebSearchTool, add 'image' to search_content_types and configure image_settings.max_results to retrieve image results alongside text; URLs and thumbnails are available in RunResult.raw_responses.
With FileSearchTool, set max_num_results between 1 and 50 as needed: leaving it at None or zero delegates the choice to the provider.
When you have many tools, group them into namespaces using tool_namespace() and add a single ToolSearchTool(): the model loads only the subset it needs each turn, reducing token consumption.
For ProgrammaticToolCallingTool, decorate each function with allowed_callers=['programmatic'] or ['direct', 'programmatic'] depending on whether you want to allow only the generated program or also direct model calls.
To reuse an existing container across runs with ShellTool, pass environment={'type': 'container_reference', 'container_id': 'cntr_...'} instead of container_auto.

Watch out

ProgrammaticToolCallingTool and ToolSearchTool only work with OpenAI Responses models: they are rejected by Chat Completions models and non-Responses backends.
With ProgrammaticToolCallingTool, if a tool returns a structured type (Pydantic, TypedDict), the default failure formatter is disabled: a handler exception propagates unless you provide a failure_error_function returning schema-conforming JSON.

In practice

Price: Free
Commercial use: Commercial use: yes
Available on: sdk-python, sdk-ts
Host: openai
Licence: MIT
FR interface: No

Security and data

Rating: Security: one caveat, see the card
Capabilities: readswritesexecutesnetwork
Your data: Data with the vendor
Maintenance: Active

Links and repository checked on 25 October 2026 · added on 1 March 2025

How to install Tools OpenAI Agents SDK

1. Make sure Python is installed (version compatible with openai>=2.25.0 for ToolSearchTool). 2. Install the SDK: pip install openai-agents 3. Export your OpenAI API key as an environment variable (OPENAI_API_KEY) rather than hardcoding it in your files. 4. Import the desired tools: from agents import Agent, WebSearchTool, FileSearchTool, Runner 5. Configure your agent by passing tools in the tools=[] parameter and run it with Runner.run() or Runner.run_sync().

pip install openai-agents

Official documentation ↗

In the same category

Whole category Orchestrate →