Build AI voice agents, control live calls over WebSocket, and manage every SignalWire resource over REST -- all from one crate.
| Capability | What it does | Quick link |
|---|---|---|
| AI Agents | Build voice agents that handle calls autonomously -- the platform runs the AI pipeline, your code defines the persona, tools, and call flow | Agent Guide |
| RELAY Client | Control live calls and SMS/MMS in real time over WebSocket -- answer, play, record, collect DTMF, conference, transfer, and more | RELAY docs |
| REST Client | Manage SignalWire resources over HTTP -- phone numbers, SIP endpoints, Fabric AI agents, video rooms, messaging, and 18+ API namespaces | REST docs |
cargo add signalwireEach agent is a self-contained microservice that generates SWML (SignalWire Markup Language) and handles SWAIG (SignalWire AI Gateway) tool calls. The SignalWire platform runs the entire AI pipeline (STT, LLM, TTS) -- your agent just defines the behavior.
use signalwire::agent::AgentBase;
use signalwire::swaig::FunctionResult;
use std::collections::HashMap;
fn main() {
let agent = AgentBase::builder("my-agent", "/agent")
.add_language("English", "en-US", "rime.spore")
.prompt_add_section("Role", "You are a helpful assistant.")
.define_tool("get_time", "Get the current time", |_args, _raw| {
let now = chrono::Local::now().format("%H:%M:%S");
FunctionResult::new(format!("The time is {now}"))
})
.build();
agent.run();
}Test locally without running a server:
cargo run --bin swaig-test -- --list-tools examples/simple_agent.rs
cargo run --bin swaig-test -- --dump-swml examples/simple_agent.rs
cargo run --bin swaig-test -- --exec get_time examples/simple_agent.rs- Prompt Object Model (POM) -- structured prompt composition via
prompt_add_section() - SWAIG tools -- define functions with
define_tool()that the AI calls mid-conversation, with native access to the call's media stack - Skills system -- add capabilities with one-liners:
agent.add_skill("datetime", None) - Contexts and steps -- structured multi-step workflows with navigation control
- DataMap tools -- tools that execute on SignalWire's servers, calling REST APIs without your own webhook
- Dynamic configuration -- per-request agent customization for multi-tenant deployments
- Call flow control -- pre-answer, post-answer, and post-AI verb insertion
- Prefab agents -- ready-to-use archetypes (InfoGatherer, Survey, FAQ, Receptionist, Concierge)
- Multi-agent hosting -- serve multiple agents on a single server with
AgentServer - SIP routing -- route SIP calls to agents based on usernames
- Session state -- persistent conversation state with global data and post-prompt summaries
- Security -- auto-generated basic auth, function-specific HMAC tokens, SSL support
- Serverless -- deploy to Lambda, Cloud Functions, Azure Functions
The examples/ directory contains working examples:
| Example | What it demonstrates |
|---|---|
| simple_agent.rs | POM prompts, SWAIG tools, multilingual support, LLM tuning |
| contexts_demo.rs | Multi-persona workflow with context switching and step navigation |
| data_map_demo.rs | Server-side API tools without webhooks |
| skills_demo.rs | Loading built-in skills (datetime, math) |
| call_flow_and_actions_demo.rs | Call flow verbs, debug events, FunctionResult actions |
| session_and_state_demo.rs | OnSummary, global data, post-prompt summaries |
| multi_agent_server.rs | Multiple agents on one server |
| lambda_agent.rs | AWS Lambda deployment |
| comprehensive_dynamic_agent.rs | Per-request dynamic configuration, multi-tenant routing |
See examples/README.md for the full list organized by category.
Real-time call control and messaging over WebSocket. The RELAY client connects to SignalWire via the Blade protocol and gives you async, imperative control over live phone calls and SMS/MMS.
use signalwire::relay::RelayClient;
use std::env;
#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
let client = RelayClient::builder()
.project(&env::var("SIGNALWIRE_PROJECT_ID")?)
.token(&env::var("SIGNALWIRE_API_TOKEN")?)
.space(&env::var("SIGNALWIRE_SPACE")?)
.contexts(vec!["default".into()])
.build()?;
client.on_call(|call| async move {
call.answer().await?;
let action = call.play(vec![serde_json::json!({
"type": "tts",
"text": "Welcome to SignalWire!"
})]).await?;
action.wait().await?;
call.hangup().await?;
Ok(())
});
println!("Waiting for inbound calls ...");
client.run().await?;
Ok(())
}- 57+ calling methods (play, record, collect, detect, tap, stream, AI, conferencing, and more)
- SMS/MMS messaging with delivery tracking
- Action objects with
wait(),stop(),pause(),resume() - Auto-reconnect with exponential backoff
See the RELAY documentation for the full guide, API reference, and examples.
Async REST client for managing SignalWire resources and controlling calls over HTTP. No WebSocket required.
use signalwire::rest::RestClient;
#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
let client = RestClient::from_env()?;
client.fabric().ai_agents().create(serde_json::json!({
"name": "Support Bot",
"prompt": {"text": "You are helpful."}
})).await?;
client.calling().dial(serde_json::json!({
"from": "+15559876543",
"to": "+15551234567",
"url": "https://example.com/call-handler"
})).await?;
let results = client.phone_numbers().search(
&[("area_code", "512")]
).await?;
println!("{results:#?}");
Ok(())
}- 21 namespaced API surfaces: Fabric (13 resource types), Calling (37 commands), Video, Datasphere, Compat (Twilio-compatible), Phone Numbers, SIP, Queues, Recordings, and more
- Connection pooling via
reqwest::Client serde_json::Valuereturns -- raw JSON, no wrapper objects
See the REST documentation for the full guide, API reference, and examples.
Add to your Cargo.toml:
[dependencies]
signalwire = "1"Or with cargo:
cargo add signalwireRequires Rust edition 2024 (rustc 1.85+).
Full reference documentation is available at developer.signalwire.com/sdks/agents-sdk.
Guides are also available in the docs/ directory:
- Agent Guide -- creating agents, prompt configuration, dynamic setup
- Architecture -- SDK architecture and core concepts
- SDK Features -- feature overview, SDK vs raw SWML comparison
- SWAIG Reference -- function results, actions, post_data lifecycle
- Contexts and Steps -- structured workflows, navigation, gather mode
- DataMap Guide -- serverless API tools without webhooks
- LLM Parameters -- temperature, top_p, barge confidence tuning
- SWML Service Guide -- low-level construction of SWML documents
- Skills System -- built-in skills and the modular framework
- Third-Party Skills -- creating and publishing custom skills
- MCP Gateway -- Model Context Protocol integration
- CLI Guide --
swaig-testcommand reference - Cloud Functions -- Lambda, Cloud Functions, Azure deployment
- Configuration -- environment variables, SSL, proxy setup
- Security -- authentication and security model
- API Reference -- complete type and method reference
- Web Service -- HTTP server and endpoint details
- Skills Parameter Schema -- skill parameter definitions
| Variable | Used by | Description |
|---|---|---|
SIGNALWIRE_PROJECT_ID |
RELAY, REST | Project identifier |
SIGNALWIRE_API_TOKEN |
RELAY, REST | API token |
SIGNALWIRE_SPACE |
RELAY, REST | Space hostname (e.g. example.signalwire.com) |
SWML_BASIC_AUTH_USER |
Agents | Basic auth username (default: auto-generated) |
SWML_BASIC_AUTH_PASSWORD |
Agents | Basic auth password (default: auto-generated) |
SWML_PROXY_URL_BASE |
Agents | Base URL when behind a reverse proxy |
SWML_SSL_ENABLED |
Agents | Enable HTTPS (true, 1, yes) |
SWML_SSL_CERT_PATH |
Agents | Path to SSL certificate |
SWML_SSL_KEY_PATH |
Agents | Path to SSL private key |
SIGNALWIRE_LOG_LEVEL |
All | Logging level (debug, info, warn, error) |
SIGNALWIRE_LOG_MODE |
All | Set to off to suppress all logging |
# Run the test suite
cargo test
# Run with verbose output
cargo test -- --nocapture
# Run tests for a specific module
cargo test logging
cargo test agent
cargo test relay
cargo test rest
# Coverage (requires cargo-tarpaulin)
cargo tarpaulin --out htmlMIT -- see LICENSE for details.