Std.Ai.Agent
Types
type alias Input = { messages : List Message }
type alias Output = { answer : String , usage : Provider.Usage }
type alias Options
type alias Journalled a = { response : a , facts : Provider.CallFacts }
type alias ToolOutput = { answer : String }
type Action
type alias Parsed = { tool : Maybe String , args : Maybe String , answer : Maybe String }
Values
defaultOptions : Options
No tracing.
inputCodec : Codec InputnativeToolLoop : String -> Provider -> String -> List Tool -> Int -> Durable.WorkflowDef Input ToolOutput
Like `toolLoop`, but uses the provider's NATIVE function-calling API
nativeToolLoopWith : Options -> String -> Provider -> String -> List Tool -> Int -> Durable.WorkflowDef Input ToolOutput
`nativeToolLoop` with `Options` (tracing).
oneShot : String -> Provider -> String -> Durable.WorkflowDef Input Output
A one-shot durable agent: prepend `systemPrompt`, call `provider` once inside
oneShotWith : Options -> String -> Provider -> String -> Durable.WorkflowDef Input Output
`oneShot` with `Options` (tracing).
outputCodec : Codec OutputstructuredToolLoop : Provider.ResponseShape a -> String -> Provider -> String -> List Tool -> Int -> Durable.WorkflowDef Input a
`nativeToolLoop` whose final answer has a shape: each model turn asks for
structuredToolLoopWith : Options -> Provider.ResponseShape a -> String -> Provider -> String -> List Tool -> Int -> Durable.WorkflowDef Input a
`structuredToolLoop` with `Options` (tracing).
toolLoop : String -> Provider -> String -> List Tool -> Int -> Durable.WorkflowDef Input ToolOutput
A tool-calling agent as a durable workflow. The model is told the `tools` and
toolLoopWith : Options -> String -> Provider -> String -> List Tool -> Int -> Durable.WorkflowDef Input ToolOutput
`toolLoop` with `Options` (tracing).