*** title: AgentBase slug: /reference/typescript/agents/agent-base description: The central class for building AI-powered voice agents with SignalWire. max-toc-depth: 3 --------------------- For a complete index of all SignalWire documentation pages, fetch https://signalwire.com/docs/llms.txt [ai]: /docs/swml/reference/ai [swml-reference]: /docs/swml/reference/ai [agentserver]: /docs/server-sdks/reference/typescript/agents/agent-server [setnativefunctions]: /docs/server-sdks/reference/typescript/agents/agent-base/native-functions [ref-datamap]: /docs/server-sdks/reference/typescript/agents/data-map `AgentBase` is the central class in the SignalWire AI Agents SDK. It provides a complete framework for building AI-powered voice agents, combining prompt management, tool definitions, skill loading, speech configuration, and web serving into a single composable interface. `AgentBase` uses **composition** internally, assembling functionality from `PromptManager`, `SwmlBuilder`, `SessionManager`, `ContextBuilder`, and `SkillManager`. All setter methods return `this` for fluent method chaining. AgentBase generates a SWML document with the [`ai`][ai] verb. See the [SWML reference][swml-reference] for the full specification of all supported parameters and behaviors. ## **Constructor** ```typescript {3} import { AgentBase } from '@signalwire/sdk'; const agent = new AgentBase({ name: 'support', route: '/support', }); ``` ### Constructor Parameters Display name of the agent. Used in logging, SIP username mapping, and the default prompt fallback. HTTP route path where this agent is served. Used by [`AgentServer`][agentserver] when hosting multiple agents on one process. Network interface the HTTP server binds to. Port the HTTP server listens on. Defaults to the `PORT` environment variable, falling back to `3000`. Explicit `[username, password]` tuple for HTTP Basic Auth on all endpoints. If not set, credentials are read from `SWML_BASIC_AUTH_USER` / `SWML_BASIC_AUTH_PASSWORD` env vars, or auto-generated on startup. Enable Prompt Object Model for structured prompt building. Set to `false` to use plain text prompts only. Expiration time in seconds for SWAIG function authentication tokens. Automatically add an `answer` verb before the AI verb in the SWML document. Enable call recording. When `true`, a `record_call` verb is added to the SWML document. Recording file format. Common values: `"mp4"`, `"wav"`. Record in stereo (separate channels for each party) when `true`. Base URL for SWAIG function webhooks. If not set, the SDK auto-detects from the incoming request or uses `SWML_PROXY_URL_BASE`. List of native SWAIG function names to enable at construction time (e.g., `["check_time", "wait_for_user"]`). Can also be set later via [`setNativeFunctions()`][setnativefunctions]. Unique identifier for this agent instance. Auto-generated as a random hex string if not provided. Suppress SDK log output. Useful in testing or when integrating with external logging. ## **Properties** The agent's display name. Set at construction time. HTTP route path where this agent is served. Network interface the HTTP server binds to. Port the HTTP server listens on. Unique identifier for this agent instance. ## **Static Members** Class-level attribute. Subclasses can set this to declaratively define prompt sections instead of calling `promptAddSection()` in the constructor. ## **Examples** ### Basic agent with a tool ```typescript {3} import { AgentBase, FunctionResult } from '@signalwire/sdk'; const agent = new AgentBase({ name: 'support-agent', route: '/support' }); agent.addLanguage({ name: 'English', code: 'en-US', voice: 'rime.spore' }); agent.setPromptText('You are a friendly customer support agent.'); agent.addHints(['SignalWire', 'SWML', 'SWAIG']); agent.setParams({ temperature: 0.7, end_of_speech_timeout: 1000 }); agent.defineTool({ name: 'check_order', description: 'Look up the status of a customer order', parameters: { type: 'object', properties: { order_id: { type: 'string', description: 'The order ID to look up' }, }, }, required: ['order_id'], handler: async (args) => { const orderId = args.order_id; return new FunctionResult(`Order ${orderId} shipped on March 28.`); }, }); await agent.run(); ``` ### Subclass with declarative prompt sections ```typescript {3} import { AgentBase, FunctionResult } from '@signalwire/sdk'; class SupportAgent extends AgentBase { static override PROMPT_SECTIONS = [ { title: 'Role', body: 'You are a customer support agent for Acme Corp.', }, { title: 'Guidelines', bullets: ['Be polite and professional', 'Escalate billing disputes'], }, ]; protected override defineTools(): void { this.defineTool({ name: 'transfer_call', description: 'Transfer the caller to a live agent', handler: async () => { return new FunctionResult('Transferring now.').connect('+15551234567'); }, }); } } const agent = new SupportAgent({ name: 'support', route: '/support' }); await agent.run(); ``` ## **Methods** Configure the answer verb that connects the call. Add a remote function include to the SWAIG configuration. Add a single speech recognition hint to improve transcription accuracy. Add multiple speech recognition hints at once. Add filler phrases for a specific function and language. Add a language configuration with voice settings for multilingual conversations. Add an external MCP server for tool discovery and invocation. Add a speech recognition hint with pattern matching and replacement. Add a SWML verb to run after the AI conversation ends. Add a SWML verb to run after the call is answered but before the AI starts. Add a SWML verb to run before the call is answered. Add a pronunciation rule to correct how the AI speaks a specific word or phrase. Load and activate a skill on the agent. Append query parameters to all SWAIG webhook URLs. Get the agent's Hono app for mounting as a sub-router. Remove all post-AI verbs from the call flow. Remove all post-answer verbs from the call flow. Remove all pre-answer verbs from the call flow. Define multi-step conversation contexts and workflows. Define a SWAIG tool that the AI can invoke during conversations. Define a tool with a typed handler and automatic schema inference. Override hook that registers tools during construction. Enable real-time debug event webhooks from the AI module during calls. Expose the agent's tools as an MCP server endpoint. Enable SIP-based call routing for this agent. Extract a SIP username from a request body (static method). Get the Hono application instance. Retrieve the agent's Basic Auth credentials and their origin. Get the full URL for this agent's endpoint. Get the list of configured MCP servers. Get the agent's display name. Retrieve the current post-prompt text. Retrieve the current rendered prompt text. Get summaries of all registered tools. Look up a registered tool by name. Handle an MCP JSON-RPC 2.0 request. Check whether a specific skill is currently loaded. Check if the MCP server endpoint is enabled. List all currently loaded skills. Manually set the proxy URL base for webhook callbacks. Set the list of native platform functions. Lifecycle hook for debug events received at /debug\_events. Pre-execution hook called before each SWAIG function invocation. Handle post-prompt summaries generated after a conversation ends. Lifecycle hook called on every SWML request before rendering. Add a new section to the agent's structured prompt. Add a subsection to an existing prompt section. Append content to an existing prompt section. Check whether a named section exists in the agent's prompt. Register a SIP username to route calls to this agent. Register a raw SWAIG function descriptor, typically from a DataMap. Unload a skill from the agent. Render the complete SWML document. Start the HTTP server (alias for serve). Start the Hono HTTP server to serve SWML and SWAIG endpoints. Set a callback for per-request dynamic agent configuration. Replace the global data object available to the AI. Replace all language configurations at once. Set a single AI parameter by key. Configure AI model parameters such as temperature and timeouts. Set the post-prompt for generating call summaries. Set LLM parameters for the post-prompt. Override the URL where post-prompt summaries are delivered. Set LLM parameters for the main prompt. Set the agent's system prompt as raw text. Register signal handlers for graceful shutdown (static method). Override the default webhook URL for SWAIG function calls. Merge data into the global data object. Custom basic-auth validation hook.