Practical overview
When it fits
Use it when an application needs a self-hosted C++ agent runtime with local-first execution, explicit tool validation and optional model connectivity. The SQLite mode is relevant when request replay and evidence-based recovery are part of the integration design.
Who it is for
- C++ developers embedding agents in edge, desktop or server software.
- Automotive and IoT teams defining local tools and deterministic commands.
- Engineers integrating local model servers with explicit tool and argument validation.
- Builders needing SQLite-backed request receipts and recovery across process restarts.
Documented uses
- Route exact phrases to deterministic skills before model inference.
- Register host-defined tools and validate their names and primitive object-schema arguments.
- Connect an optional local model server through the documented HTTP adapter.
- Persist request receipts and history with SQLite for deduplication and recovery.
- Integrate business tools that report committed, failed or unknown outcomes.
Requirements & limits
The public repository provides a CLI and installable C++ SDK. The documented build targets Linux and macOS; optional HTTP and SQLite capabilities require their respective development dependencies.
- Choose the deterministic path when exact-phrase skills can handle the request without a model.
- Use the HTTP adapter only with a separately started compatible llama-server; the CLI does not download or start a model.
- Enable SQLite explicitly when cross-restart receipts, history recovery and request deduplication are needed.
- Plan external verification for UNKNOWN outcomes because SQLite does not provide a transaction with a real device.
- Account for the supported primitive object-schema subset and the documented synchronous runtime model.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Build the CLI and run the local simulator
Compile Masteragent and run its documented automotive demo without a model, credentials, network or vehicle hardware.
Before you start
- CMake 3.18 or newer
- C++17 compiler
- libcurl and SQLite development files for the standard build
- Python 3 for tests and packaging
Steps
Clone the MasterAgent repository and change into its directory.
Configure a Release build with CMake.
Build the project and run the test command.
Run the
sparx demo automotiveexecutable.
Expected outcome
The demo reads back an in-memory simulated AC temperature and emits JSON output.
Guide limitations
- The simulator state resets when the process ends.
- The documented release CI targets Linux and macOS.
- This recipe does not connect to a real vehicle or start a model.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 3 October 2026
- Original website title
- OpenSparX/MasterAgent · Official homepage title · retained in its source language

