Module search 

Search — LLM-routed search across grep, glob, web, memory, and RLM.

The router uses codetether models discovery (via [ProviderRegistry]) plus a single LLM call to pick the best backend for a natural-language query, then invokes that backend and returns a normalized result::RouterResult.

Quick start

use std::sync::Arc;
use codetether_agent::provider::ProviderRegistry;
use codetether_agent::search::run_router_search;
use codetether_agent::search::model::DEFAULT_ROUTER_MODEL;

let registry = Arc::new(ProviderRegistry::from_vault().await.unwrap());
let res = run_router_search(registry, DEFAULT_ROUTER_MODEL, "where is fn main", 1).await.unwrap();
assert_eq!(res.runs.len(), 1);

Re-exports

pub use engine::run_router_search;

pub use result::BackendRun;

pub use result::RouterResult;

pub use types::Backend;

pub use types::BackendChoice;

Modules

dispatch

Dispatch a BackendChoice to the underlying tool implementation and return its ToolResult. Each backend is invoked inline so the router does not require access to the full ToolRegistry.

engine

End-to-end: resolve model → ask router → run chosen backends → collect.

model

Resolve a router provider/model string against the live ProviderRegistry.

parse

JSON parser for the router LLM response.

prompt

Prompt builder for the LLM-based search router.

request

Build the CompletionRequest used to ask the router model for a plan.

result

Normalized result type produced by the router and consumed by the CLI.

types

Types shared between the search router, its backends, and the CLI.