hypersteeldb 0.2.4

A database that compiles questions instead of guessing answers: typed vocabulary discovered from your documents, queries type-checked before they run, roaring-bitmap set algebra over reified hyperedges, and Dempster-Shafer evidence with an explicit conflict guard.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21

//! The provider abstraction: any chat model with tool-use, behind one async trait. Bedrock (Claude)
//! and Paddock (local Qwen) both implement it; the agent loop is provider-agnostic.

use crate::agent::types::{Msg, ToolSpec, Turn};
use async_trait::async_trait;

#[async_trait]
pub trait LlmProvider: Send + Sync {
    /// One turn: given the system prompt, conversation, and available tools, return the model's reply
    /// (assistant text and/or tool calls).
    async fn chat(&self, system: &str, msgs: &[Msg], tools: &[ToolSpec]) -> Result<Turn, String>;
    fn name(&self) -> &str;

    /// Whether this provider produces trustworthy natural-language *synthesis* of tool results. Large
    /// models (Bedrock, big OpenAI-compatible) do; small on-device tool-callers (Needle, the native
    /// 1.7B) do NOT — for those the agent renders a deterministic template from the program result
    /// instead of trusting the model to phrase the numbers. Default `true`.
    fn synthesizes(&self) -> bool {
        true
    }
}