Reference · Glossary
- The language
- Agents and tools
- Model providers
- The Anthropic API
- Talking agents
- Standard library
- Glossary
- Package docs
- Telemetry
- Robots
- Servers
- Deploying
- tin.ad
- The tin command
- How it works
On this page
Glossary
Tin’s own words, each in one sentence, with where to read more.
- agent
- A declaration with a model, a prompt and tools; calling it runs the model loop and returns its reply. Learn: your first agent
- capability
- Something a program can reach outside itself (network, files, a tool, a sub-agent); the compiler reports each agent’s before it runs. Learn: capabilities
- grant
- What a server is allowed to do, written on its declaration (for example net.listen(8080)); anything it reaches beyond that is a compile error. Servers: grants
- guarded
- A guarded tool runs only after a person approves the call, and any path that skips the approval is a compile error. Learn: guarded tools
- mock model
- A built-in model that answers offline and the same way every time, so examples and tests are repeatable. Learn: your first agent
- taint
- Text from a model or a request is tainted: it cannot reach outside code, a URL or a query until you parse it, encode it or approve it. Servers: request taint
- tape
- A recording of a run (model replies, fetches, input, random numbers) that --replay plays back with no model and no network. Learn: record and replay
- task
- spawn starts a task that runs at the same time as the code that started it; await takes its result, within gives it a deadline. Learn: spawn, await and within
- tool
- A function marked tool fn that a model may call; its JSON schema is generated from its signature and /// docs. Learn: tools
- typed reply
- Asking an agent for a type: the reply is decoded into a checked value of that type instead of text. Learn: typed replies