Dependencies

AgentContext stores typed and named dependencies. Runtime hooks and tools can read the same dependency values during a run.

Tool dependency

#![allow(unused)]
fn main() {
use std::sync::Arc;

use serde_json::json;
use starweaver_agent::{AgentBuilder, FunctionTool, TestModel, ToolContext, ToolRegistry, ToolResult};
use starweaver_context::AgentContext;

#[derive(Debug)]
struct Tenant(String);

async fn example() -> Result<(), starweaver_agent::AgentError> {
let tool = FunctionTool::new(
    "tenant",
    Some("Return tenant".to_string()),
    json!({"type": "object"}),
    |ctx: ToolContext, _args: serde_json::Value| async move {
        let tenant = ctx.dependency::<Tenant>().expect("tenant dependency");
        Ok(ToolResult::new(json!({"tenant": tenant.0})))
    },
);

let tools = ToolRegistry::new().with_tool(Arc::new(tool));
let mut context = AgentContext::default();
context.insert_dependency(Tenant("acme".to_string()));

let agent = AgentBuilder::new(Arc::new(TestModel::with_text("done")))
    .tool_registry(tools)
    .build();
let result = agent.run_with_context("hello", &mut context).await?;
assert_eq!(result.output, "done");
Ok(())
}
}

Dependencies are process-local and skipped during serialization. Service runtimes should rehydrate them from application configuration after restoring resumable state.

Notes

AgentContext also carries serializable notes for lightweight session memory. Notes round-trip through ResumableState; context instructions expose note keys while keeping note values out of model-facing prompt text.

#![allow(unused)]
fn main() {
use starweaver_context::AgentContext;

let mut context = AgentContext::default();
context.notes.set("lang", "Chinese");
context.notes.set("os", "macOS");

let restored = AgentContext::from_state(context.export_state());
assert_eq!(restored.notes.get("lang"), Some("Chinese"));

let instructions = restored.context_instructions(true).unwrap();
assert!(instructions.contains("key=\"lang\""));
assert!(!instructions.contains("Chinese"));
}