Documentation
¶
Overview ¶
Package agent implements the agent loop and MCP client for Orla Agent Mode (RFC 4).
Package agent implements the agent loop and MCP client for Orla Agent Mode (RFC 4).
Package agent implements the agent loop and MCP client for Orla Agent Mode (RFC 4).
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func ExecuteAgentPrompt ¶
func ExecuteAgentPrompt(prompt string, modelOverride string, daemonURL string, profileName string, workflowName string) error
ExecuteAgentPrompt is the main entry point for agent execution It handles the full flow: config loading, executor creation, context/signal handling, and execution prompt: the agent prompt as a single string (should be quoted when called from CLI) modelOverride: optional model override (for backward compatibility) daemonURL: optional daemon URL to connect to (RFC 5) profileName: optional agent profile name from config (RFC 5) workflowName: optional workflow name to execute (RFC 5)
Types ¶
type Client ¶
Client wraps an MCP client session for communicating with the internal Orla server
func NewClient ¶
NewClient creates a new MCP client that connects to the internal Orla server via stdio The server is started as a subprocess running "orla serve --stdio" If orlaBin is empty, it will use the current executable
func (*Client) CallTool ¶
func (c *Client) CallTool(ctx context.Context, params *mcp.CallToolParams) (*mcp.CallToolResult, error)
CallTool calls a tool via the MCP server
type Executor ¶
type Executor struct {
// contains filtered or unexported fields
}
Executor handles agent execution with proper setup and teardown
func NewExecutor ¶
func NewExecutor(cfg *config.OrlaConfig) (*Executor, error)
NewExecutor creates a new agent executor
type Loop ¶
type Loop struct {
// contains filtered or unexported fields
}
Loop orchestrates the agent execution flow
func (*Loop) Execute ¶
func (l *Loop) Execute(ctx context.Context, prompt string, messages []model.Message, stream bool, streamHandler StreamHandler) (*model.Response, error)
Execute runs a single agent execution cycle It implements the agent loop from RFC 4 Section 4.4: 1. Receive user prompt 2. Send prompt and available tools to the model 3. Receive model response (may include tool calls) 4. Execute tool calls via MCP 5. Return tool results to the model 6. Receive final response from the model 7. Stream response to user (if streaming and handler provided)
If streamHandler is provided and streaming is enabled, it will be called for each chunk. The stream will be consumed before checking for tool calls, ensuring the response is complete.
type MCPClient ¶
type MCPClient interface {
ListTools(ctx context.Context) ([]*mcp.Tool, error)
CallTool(ctx context.Context, params *mcp.CallToolParams) (*mcp.CallToolResult, error)
}
MCPClient is an interface for MCP client operations used by the agent loop
type StreamHandler ¶
type StreamHandler func(event model.StreamEvent) error
StreamHandler is a function that handles streaming events
Source Files