Concepts

Providers and model routing

Vesper natively supports OpenAI, Anthropic (Claude), and Google (Gemini). The execution engine routes your request to the correct provider API based on the model string you define in your manifest (for example gpt-4o-mini, claude-3-5-sonnet, or gemini-1.5-pro). You never write custom provider wrappers.

Tools

Tools give your agents the ability to interact with the outside world. Vesper provides built-in tools like web_search and read_file, and lets you attach custom Python functions with the @vesper.tool decorator in your entryPoint file.

Memory scopes and sessions

Vesper's Memory Engine provides persistent, stateful memory with automatic retention windows and overflow truncation. Configure the scope as per-session (cleared when the session ends) or per-project (shared across all runs in the project).

FinOps and budgets

Cost governance is a first-class citizen. By defining maxCostPerRun and alertAt in your manifest, the FinOps engine tracks the cost of every run in real time, enforces hard budget caps to prevent runaway token usage, and logs the exact cost of every session in the local registry.