Documentation
¶
Overview ¶
Package model contains the core data types shared across react-agent and its sub-packages.
The types here are intentionally minimal and framework-agnostic so that the parent package imports only what it needs, and external adapters can convert to/from their own representations without depending on the orchestration logic.
Key types ¶
- ToolDefinition — JSON-schema description of a function the LLM can call.
- ToolExecutor — interface for dispatching ToolCall slices to a backend.
- ContentItem — sealed discriminated union: Message, ToolCall, ToolResult.
- Event — a timestamped, authored entry in an [ExecutionContext] history.
- Request / Response — the exchange contract between Agent and [LLMClient].
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
This section is empty.
Types ¶
type ContentItem ¶
type ContentItem interface {
Type() string
// contains filtered or unexported methods
}
ContentItem is a closed discriminated union of all values that can appear in an Event. The unexported marker method prevents external packages from satisfying the interface, keeping type switches exhaustive.
type Event ¶
type Event struct {
ID string `json:"id"`
ExecutionID string `json:"execution_id"`
Timestamp time.Time `json:"timestamp"`
Author string `json:"author"` // "user" | "agent" | "tools"
Content []ContentItem `json:"content"`
}
Event is the atomic unit of history in an ExecutionContext. Author identifies the source: "user", "agent", or "tools". One event per step-participant — the LLM produces one agent event per Think, the tool runner produces one tools event per Act.
type Request ¶
type Request struct {
Instructions string
Events []Event
Tools []ToolDefinition
MaxTokens int64
}
Request is the agent's view of a single LLM call. Events carries the full conversation history; the LLMClient translates each Event into the provider-specific message format.
type Response ¶
type Response struct {
Content []ContentItem
}
Response is the agent's view of a single LLM reply. Content contains either ToolCall items (when the model wants to act) or a single Message item with Role "assistant" (when the model has an answer).
type ToolCall ¶
type ToolCall struct {
ID string `json:"tool_call_id"`
Name string `json:"name"`
Arguments json.RawMessage `json:"arguments"`
}
ToolCall is a single tool invocation requested by the LLM.
type ToolDefinition ¶
type ToolDefinition struct {
Name string `json:"name"`
Description string `json:"description"`
// Parameters is a JSON Schema object describing the function arguments.
// Example: map[string]any{"type": "object", "properties": {...}, "required": [...]}
Parameters map[string]any `json:"parameters"`
Strict bool `json:"strict,omitempty"`
}
ToolDefinition describes a function the LLM can invoke. This type is intentionally independent of any specific tool framework — convert from mcp-toolkit, langchain, or your own registry with a simple loop.
type ToolExecutor ¶
type ToolExecutor interface {
Execute(ctx context.Context, calls []ToolCall) ([]ToolResult, error)
}
ToolExecutor abstracts concurrent tool dispatch for the Agent. Implement this interface to connect any tool-running backend. The default adapter for mcp-toolkit is in the mcpadapter subpackage.
type ToolResult ¶
type ToolResult struct {
ID string `json:"tool_call_id"`
Name string `json:"name"`
Status string `json:"status"` // "success" | "error"
Content []string `json:"content"`
}
ToolResult is the outcome of executing a ToolCall.
func (ToolResult) Type ¶
func (ToolResult) Type() string
Source Files