ENGINEERING TOOLS
Tai-IoT engineering tools
Describe the task in natural language and your AI client picks the right engineering tools to read, change, build, test and review. You never need to remember tool names.
What the tools do
The tools cover six kinds of engineering work. For changes and device operations, the client can ask you to confirm before anything runs.
Project reading
Learn the target model, project configuration and source structure, and confirm the context before any change. Read-only; the project is not modified.
Project changes
Change configuration or source in the working copy only, leaving a reviewable diff for every change.
Build and test
Build and run tests as the project requires and return the full output. Tests never run on a build older than the latest change, so stale results cannot mislead.
Version diff
Summarize the changes for the AI to explain and the engineer to review. Never commits or pushes on its own.
Device operations
Download and flash test programs, provided with support packs. Enabled only for designated test devices, with engineer confirmation before each run.
Runtime data
View device status and diagnostics, provided with support packs. Read-only and opened within the authorized scope.
Host example tools
The host example in the free trial toolkit provides these 6 tools, which complete a full engineering task on the included temperature acquisition C example project.
| Action | Description | Type |
|---|---|---|
View project overviewget_project_profile | Target, sampling parameters, file list and available operations. | Project reading |
Read project filesread_project_file | Reads the example project's configuration and C source; only files inside the example project are accessible. | Project reading |
Change sampling parametersset_sampling | Changes the sampling period and alarm threshold. Values outside the allowed range are refused and the configuration stays unchanged. | Project changes |
Build the projectbuild_project | Builds for real on your computer and returns the full compiler output. | Build and test |
Run simulated testsrun_simulation | Runs 7 checks with a simulated clock and temperature input and returns 10 samples. Does not run before a build, or after a change until the project is rebuilt. | Build and test |
View the diffget_git_diff | Lists the changes made in this task; never commits or pushes. | Version diff |
These tools were tested with natural-language tasks on Codex CLI 0.162.1 (with the 0.1.1 tool server); the model chose and called all 6 tools by itself.
Development board example tools
The development board example provides these 7 tools, which change, cross-compile, flash and read back the example projects for the ESP32-S3-DevKitC-1, NUCLEO-H723ZG and AT-START-F413. Flashing and read-back are enabled only for the one board whose serial port was designated at start, and before writing they check that the build is current and its files are unchanged.
| Action | Description | Type |
|---|---|---|
View board and projectget_board_profile | Board, target chip, sampling parameters, whether the vendor SDK is ready, build state, and the USB serial devices connected. | Project reading |
Read project filesread_project_file | Reads the example project's configuration and source; only files inside the project are accessible. | Project reading |
Change sampling parametersset_sampling | Changes the sampling period and alarm threshold. Values outside the allowed range are refused and the configuration stays unchanged. | Project changes |
Cross-compile firmwarebuild_firmware | Builds with the board's pinned vendor SDK and compiler, and returns the firmware size, memory use, SHA-256 and the values the compiler actually received. | Build and test |
Flash firmwareflash_firmware | Writes the current build to the designated board; refused when the build is out of date, its files have changed or the port is not that board. | Device operations |
Read back serial outputread_serial_log | Restarts the designated board, reads its serial port for 3 to 60 seconds and checks the start settings, sampling interval, alarm flags and temperature range. | Runtime data |
View the diffget_git_diff | Lists the changes made in this task; never commits or pushes. | Version diff |
Cross-compilation for all three boards was tested through these tools on Linux; the values the compiler received match the changed configuration.
Try saying
Type the task straight into a connected AI client. The recorded runs behind these results gave the same tasks in Chinese.
Using the Tai-IoT tools, first read the project, then change the sampling period from 1000 ms to 500 ms and the alarm threshold to 28 °C. Build it for real, run the simulated test and show the Git diff. Report the actual output and do not commit or push.
Recorded result: sampling period changed to 500 ms and threshold to 28 °C, 7 checks passed, 10 samples, 3 alarm samples.
First try setting the sampling period to 55 ms with a 28 °C threshold, record what the tool returns and confirm the configuration. Then run the simulation before building and explain the error that comes back.
Recorded result: 55 ms was refused and the configuration stayed unchanged, the simulation without a build was refused, and the AI explained the cause from the tool output and recovered.
Data and where it runs
- Runs locally: the tools run on your computer, and project code never passes through Tai-IoT servers.
- No uploads: the toolkit uploads nothing and never reads your accounts or keys.
- Scope: works only on the designated project; never commits, pushes or accesses remote repositories.
- Model data: when you use a cloud model, the client sends project content and tool results to the model service you choose; enterprises can choose the model and what is sent.
Supported clients
Works with AI clients that support the MCP standard, using your existing accounts and model services.
| Client | Setup | Notes |
|---|---|---|
| Codex CLI | One registration command | Tested: 0.162.1 / gpt-6.1-sol / Linux, covering a normal task, error recovery and compile-failure handling. |
| VS Code | Workspace configuration generated by the board example | Use your own model service in the VS Code AI chat, without a GitHub sign-in. |
| Claude Code | Configuration generated by the toolkit | The toolkit generates the configuration and provides setup steps. |
| Cursor, VS Code + Cline and others | Register in client settings | Register the tools in the client's MCP settings. |