sgr-agent-core

Core types for the sgr-agent ecosystem — the minimal shared interface that both the framework and tool crates depend on.

What's inside

When to use

• Building an agent app? Use sgr-agent — it re-exports everything from core.
• Building a tools crate? Depend on sgr-agent-core to avoid circular deps with sgr-agent.

Creating a tool

use sgr_agent_core::{Tool, ToolOutput, ToolError, parse_args, AgentContext, json_schema_for};
use schemars::JsonSchema;
use serde::Deserialize;

#[derive(Deserialize, JsonSchema)]
struct PingArgs {
    /// Host to ping
    host: String,
}

struct PingTool;

#[async_trait::async_trait]
impl Tool for PingTool {
    fn name(&self) -> &str { "ping" }
    fn description(&self) -> &str { "Ping a host and return latency" }
    fn is_read_only(&self) -> bool { true }
    fn parameters_schema(&self) -> serde_json::Value { json_schema_for::<PingArgs>() }

    async fn execute(
        &self,
        args: serde_json::Value,
        _ctx: &mut AgentContext,
    ) -> Result<ToolOutput, ToolError> {
        let a: PingArgs = parse_args(&args)?;
        // ... ping logic
        Ok(ToolOutput::text(format!("{}: 12ms", a.host)))
    }
}

Crate graph

sgr-agent-core (this crate)  ← 5 lightweight deps
    ↑              ↑
sgr-agent-tools    sgr-agent
(file tools)       (framework)

Dependencies

Only 5 lightweight crates — no reqwest, no tokio, no LLM clients:

• async-trait — async trait methods
• serde + serde_json — JSON serialization
• schemars — JSON Schema generation
• thiserror — error derives