Getting started
Launch the local workbench, connect Valcore to the Gateway and Logfire, then give your coding agent the Valcore skill.
Launch Valcore
Start here after installing Valcore:
valcore serveValcore starts at http://127.0.0.1:8000 and opens the visual workbench in your browser. The UI and CLI share the same local SQLite workspace, so anything you create in one is available in the other.
To use another port or start without opening a browser:
valcore serve --port 8080
valcore serve --no-browserConfiguring keys
Open Settings in the workbench or use the CLI. Valcore stores credentials in ~/.valcore/config.toml with restricted file permissions and masks them in the UI.
Valcore project keys
Gateway key
Required to run or generate with models. Create it in Pydantic AI Gateway, then configure it:
valcore config set-keyTracing key
Optional Logfire write token for Valcore's FastAPI, Gateway, and run spans.
valcore config set-logfire-tokenWrite datasets key
Pushes curated datasets to the Valcore Logfire project. Grant project:read_datasets and project:write_datasets.
valcore config set-logfire-write-keyAgent project keys
The read key belongs to the Logfire project where your agent runs. Grant project:read and project:read_datasets so Valcore can query traces and fetch hosted datasets.
valcore config set-logfire-read-keyLocal agent
Using the CLI agent in Valcore
Install Valcore's bundled skill in the repository where your coding agent works:
cd your-agent-project
valcore skills installThe default installs to .agents/skills/. Use --claude, --copilot, --all, or --global when you need a different destination.
The app and CLI use the same local SQLite workspace, so no Valcore server needs to be running for agent-driven CLI work. Resources can be addressed by name or a unique ID prefix.
valcore datasets list
valcore evaluators list
valcore runs list