Concepts & guidesΒΆ
These guides set out the ideas behind Kaval.AI and how its parts fit together. They complement the practical Tutorials: a guide explains why a mechanism exists, a notebook demonstrates how it is used.
Readers new to building with language models should begin with Core concepts. For the design of the library as a whole, see Architecture.
- Core concepts
- Workflows
- Tools
- Agents
- Observability
- What a run records
- Persistence and logging
- Local vs. production databases
- AgentService
- Why there is no cost column
- The trajectory: what the run actually did
- Reading a trajectory without a database
- Marking non-production traffic
- TaskLogger and fire-and-forget
- Recording less
- Custom loggers and composition
- Cost per run and per conversation
- Retention
- The backoffice UI
- Evaluation & acceptance testing
- The database and data model
- Safety