Practical overview
When it fits
Use Agents-SDK when you are developing a C++ application and want to define agent planning, workflow composition and tool access within that application.
Who it is for
- C++ developers building local-first or on-device agent applications.
- Teams that want to compose agent workflows and register tools in their own software.
Documented uses
- Build a C++ agent that breaks a task into steps and uses tools to answer it.
- Compose workflows such as routing, prompt chaining, parallelization or orchestrator-workers.
- Choose among documented agent strategies, including ReAct, Plan and Execute, and Zero-Shot.
- Connect an agent to listed hosted providers or local-model options.
Requirements & limits
- Building requires a C++20-compatible compiler and Bazel 8.3.1 or later.
- The README lists OpenAI, Anthropic, Google, Ollama and llama.cpp model options; hosted providers require their corresponding API credentials.
- Brave Search credentials are required for the documented web-search tool.
- Python 3 in PATH is optional and is needed for the Python execution tool.
- Chain-of-Thought and Reflexion are marked in development in the README.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Build the SDK and configure provider keys
Prepare a local checkout for building the SDK and configuring documented model or web-search credentials.
Before you start
- C++20-compatible compiler: GCC 14+, Clang 17+ or MSVC 2022+
- Bazel 8.3.1 or later
- Provider API key for a hosted model, and Brave Search API key if using the web-search tool
Steps
Clone the repository with the command below and enter the
agents-sdkdirectory.git clone https://github.com/RunEdgeAI/agents-sdk.gitBuild the project with
bazel build ....Configure keys in a
.envfile copied from.env.template, as environment variables, or as command-line arguments.
Expected outcome
A checked-out SDK with documented build instructions and credentials configured for the providers or search tool being used.
Guide limitations
- The README's build command does not specify a target.
- Command-line API-key arguments are described as not recommended for production.
- Python 3 in PATH is only required when using the Python execution tool.
Run the simple-agent example
Use the documented example to submit a task to an autonomous agent with a selected model and registered web-search tool.
Before you start
- A built or buildable checkout, C++20-compatible compiler and Bazel 8.3.1 or later
- An API key for the selected hosted model
- A Brave Search API key when using the web-search tool
Steps
Configure the relevant provider and search credentials using the documented
.env, environment-variable or command-line options.From the
agents-sdkdirectory, runbazel run examples:simple_agent -- your_api_key_here.Enter a question or task when prompted; approve a proposed step when the example requests approval.
Expected outcome
The example agent can plan task steps, use tools such as web search to gather information and return an answer.
Guide limitations
- The README's example command shows a command-line API key; its alternative environment-variable example also includes a command-line argument.
- Approval prompts depend on human-in-the-loop being enabled.
- The example's provider and model must be configured for the credentials supplied.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 18 October 2026
- Original website title
- RunEdgeAI/agents.cpp · Official homepage title · retained in its source language